The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC07)

MATCHED FILTER TIME AND FREQUENCY SYNCHRONIZATION METHOD FOR OFDM SYSTEMS USING PN-SEQUENCE PREAMBLES
Henri Puska and Harri Saarnisaari University of Oulu, Centre for Wireless Communications (CWC) P.O.Box 4500, FI-90014 Oulu, FINLAND A BSTRACT Matched lter (MF) time and frequency synchronization method for orthogonal frequency division multiplexing (OFDM) systems is presented. The proposed method uses the same pseudo-noise (PN)-preamble generation as was used in [1], but signal processing in the receiver is different. The performance of the proposed method is investigated using theoretical and simulated synchronization probabilities. The obtained results indicate that the proposed synchronizer attains the optimum theoretical performance. I. I NTRODUCTION proposal utilizes matched lter (MF) for time and frequency synchronization. The rest of the paper is organized as follows. The proposed time and frequency synchronization methods are described in Sections II and III, respectively. In Section IV simulated results are compared with theoretical ones in additive white Gaussian noise (AWGN) channel for verifying obtained results. The results are also compared to the Tufvessons method [8], where time domain PN-preamble and slightly different timing metric are used. Finally, conclusions are drawn in Section V. II. In orthogonal frequency division multiplexing (OFDM) systems, time and frequency synchronization are important issues to be solved [2]. If these tasks are not performed with sufcient accuracy, the orthogonality among the sub-carriers is lost, and the communication system suffers from intersymbol interference (ISI) and intercarrier interference (ICI). Several techniques have been proposed recently for OFDM timing and frequency synchronization. Those suggested in [3], [4] propose to use the correlation that exists between the samples of the cyclic prex and the corresponding portion of the OFDM symbol. Since received cyclic prex is usually affected by ISI, the performance of this estimator depends on the channel. To avoid this problem in timing synchronization, the method in [1] uses a training symbol with two identical halves in time domain. However, the timing metric used in [1] suffers from broad autocorrelation function (ACF), which may result large timing estimation error. This blurry ACF can be eliminated and, hence, timing offset estimation error can be reduced by designing a special training symbols which give more sharper timing metric trajectory [5], [6]. It is well known that pseudo-noise (PN)-sequences have good autocorrelation properties. Therefore, many authors have proposed the use of time domain PN-sequence as preamble in OFDM systems [7], [8]. The method used in [8] is also applied in [9] but there the principle of frequency domain PN synchronizer is also presented. In this paper, we propose a method which uses the Schmidl and Coxs [1] training symbol generation. Therein, two identical halves in time domain are generated by transmitting PNsequence on the even frequencies, while zeros are transmitted on the odd frequencies. As a difference, instead of correlating two identical halves with each other in the receiver [1], our
This research was supported by the URANUS (U niversal RAdio-link Platform for EfcieN t U ser-centric AccesS) project of the EUs 6th framework programme (Project No. IST-27960 (URANUS))

P ROPOSED T IME S YNCHRONIZATION M ETHOD

A block diagram of the typical OFDM system is shown in Fig. 1. Therein, input data bits are rst encoded, interleaved and mapped to data symbols, each of which modulates a different subcarrier. In the proposed scheme, the transmitter transmits every now and then a PN-preamble, whose length is L. This PN-sequence is then zero padded such that zeros are placed on the odd frequencies after serial-to-parallel (S/P) conversion. Therefore, the total length of the PN-preamble is 2L, which is equal to the number of subcarriers (Nc ). Multicarrier modulation is performed by taking the inverse fast Fourier transform (IFFT) of data symbols. The discrete time baseband OFDM signal can be expressed as 1 s(n) = Nc
Nc 1 k=0

D(k )ej 2kn/Nc ,

(1)

where D(k ) is a data symbol which modulates k th subcarrier. When the number of subcarriers is sufciently large, s(n) approximates a complex Gaussian process with zero mean and 2 = E {|s(n)|2 } [4]. An OFDM symbol has a usevariance s ful symbol period Tu and preceding each symbol is a cyclic prex of length Tcp , which is longer than the channel impulse response so that there will be no ISI [1]. The baseband signal is digital-to-analog (D/A) transformed and up-converted to the radio frequency (RF) and then transmitted through the channel. At the receiver, discrete time baseband signal r(n) is obtained by down-converting and analog-to-digital (A/D) transforming the continuous time RF-signal. In an AWGN channel, without frequency offset, r(n) is given by r(n) = s(n )ej + (n), (2)

where is the propagation delay, is the carrier phase and 2 (n) is AWG noise with zero mean and variance = 2 E {| (n)| }. Cyclic prex can not be removed until the receiver has knowledge about frequency, phase, and symbol synchronization (Fig. 1). In this section, we concentrate on timing

1-4244-1144-0/07/$25.00 c 2007 IEEE

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC07)

Data in

Coding + interleaving

Symbol mapping

PNpreamble addition

S/P

IFFT

P/S

0.35 SNR = 0 dB

A/D r(n)

RF RX Channel RF TX D/A Cyclix prefix addition

0.3

Nc = 256 L = 128 TCP = 0

Data FFT P/S Symbol demapping Deinterleaving + decoding out

abs(MF output)

1111111 0000000 0000000 1111111 1111111 0000000

Synchronization Cyclix prefix removal S/P

0.25

0.2

L=T /2
u

0.15

Figure 1: A block diagram of the OFDM system. synchronization where the starting point of the OFDM symbol is estimated. After cyclic prex removal, multicarrier demodulation is performed by taking the fast Fourier transform (FFT). Then the received information symbols are demapped to bits which are deinterleaved and decoded. A. Time Domain MF Synchronizer Fig. 2 shows the block diagram of the noncoherent MF timing synchronizer implemented in time domain. The sampled signal r(n), from A/D converter (Fig. 1), has signal-to-noise-ratio 2 2 / . (SNR) = s
r(n) Matched filter (MF) y(n) 2 PDI z(n) sync

0.1

Angle separation () between those spikes is proportional to frequency offset (f)

0.05

50

100 150 Sample index

200

255

Figure 3: Magnitude squared output of the MF. from which the detection threshold TH can be calculated in order to obtain the desired PF A . In the synchronous code phase position decision variable z (n) is characterized by the noncentral chi-square distribution and the probability of detection (PD ) is PD = Q2
2 4Ls , 2

threshold

2TH . 2 L

(4)

Figure 2: Noncoherent MF time synchronization unit in time domain. The signal r(n) is then passed through MF whose impulse response h(n) is complex conjugated, time reversed and delayed version of one half of the sP N (n), i.e., h(n) = s P N (L n). When the preamble signal sP N (n) is ltered using this MF, then two identical sharp ACFs occur as can be seen from Fig. 3. This is because the transmitted preamble consisted of two identical halves. Post detection integration (PDI) can be used to combine these two ACFs and time synchronization is made from this combined signal shown in Fig. 4. As a reference, timing metric of the Schmidl and Coxs method [1] is shown in Fig. 5. By comparing these gures, it can be seen that the proposed scheme provides sharper timing metric. Next, the theoretical synchronization probabilities for the proposed method are presented. When sP N (n) and h(n) are not synchronized, then the output of the MF (y (n)) is (if an ideal ACF is assumed) complex valued Gaussian distributed 2 /2 random variable with zero mean and variance equal to L per I and Q branch. Therefore, the decision variable z (n) after magnitude squaring and PDI follows the central chi-square distribution with 4 degrees of freedom and the probability of false alarm (PF A ) is PF A TH = exp 2 L
1 k=0

where Q() is the generalized Marcums Q-function [10]. When the decision is made according to the maximum selection, then the largest sample from L possible samples is selected from the set {z (n)} = [z (n) z (n + 1) z (n + L 1)]. The selected largest sample can be taken either from the correct (z (c)) or from the wrong code phase (z (w)). The probability of selecting the correct sample from the set {z (n)} is denoted by PM ax = P (z (c) > z (w)i , i), where i refers to all of the L 1 possible false code phases. The probability PM ax is obtained as [11] PM ax = P (z (w)i < z (c), i) =

p(z (c))[P (z (w) < z (c))]L1 dz (c),

(5)

where p(z (c)) is the probability density function (PDF) of the random variable z (c). The probability that the maximum sample is taken from the correct time instant and, in addition, it also exceeds the threshold TH can be calculated as PD,M ax = PM ax PD , (6)

which is a useful probability when both maximum selection and threshold comparison are utilized. B. Frequency Domain MF Synchronizer

1 TH 2 k ! L

, TH 0.

(3)

Fig. 6 illustrates the frequency domain implementation of the equivalent time domain MF synchronizer structure (Fig.

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC07)

0.25 SNR = 0 dB

0.2 0.18 N = 256 SNR = 0 dB

Timing metric of the proposed method

0.2

0.16 Timing metric of the S&C algorithm 0.14 0.12 0.1 0.08 0.06 0.04 0.02

N = 256
c

L = 128 TCP = 0

L = 128

0.15

TCP = 0

0.1

0.05

20

40

60 80 Sample index

100

127

50

100 150 Sample index

200

255

Figure 4: Noncoherently combined autocorrelation function. 2). As is well known, convolution in time domain corresponds multiplication in frequency domain. Because OFDM systems already have IFFT/FFT blocks for multicarrier modulation and demodulation, it may be advantageous to use them also in matched ltering. When the MF is implemented in frequency domain, the incoming signal r(n), from A/D converter (Fig. 1), has to be transformed via FFT to the frequency domain (R(f ) = FFT[r(n)]). Impulse response h(n) of the MF has to be also transformed to frequency domain (H (f ) = FFT[h(n)]). Then R(f ) and H (f ) are multiplied and the result of this multiplication is IFFT transformed back to time domain. Remaining part of Fig.6 is exactly the same as was described in Fig. 2. It is explained, e.g., in [9], how this ltering can be performed via the overlap-save method in frequency domain. III. F REQUENCY O FFSET E STIMATION

Figure 5: Timing metric of the Schmidl and Coxs method [1].

r(n) S/P FFT Multiply by H(f) y(n) IFFT P/S 2 PDI z(n) sync

Threshold

Figure 6: Noncoherent MF time synchronization unit in frequency domain. IV. R ESULTS

A carrier frequency offset f causes a phase rotation of [1] = f Tu (7) between the two halves of the MF output signal y (n). The phase difference can be estimated from the y (n) (Fig. 3) after timing synchronization by = angle(y ( )y ( + Tu )), 2 (8)

where is the timing estimate. If || < , then the frequency offset estimate is [1] f = /(Tu ). (9)

The Cramer-Rao bound (CRB) for the normalized frequency offset can be expressed as [1] Var[f Tu ] = Var
2 = 2 2. Ls

(10)

Timing synchronization performance of the proposed method is rst evaluated. Therefore, theoretical optimum synchronization probabilities from (3), (4), (5) and (6) are calculated when L = 128, Nc = 256, f = 0 and Tcp = 0. The channel is AWGN throughout this study. Obtained theoretical probabilities are then veried by simulations. As a reference, these results are also compared to those given by the Schmidl and Coxs (S&C) [1] and the Tufvessons methods (Tufv.) [8]. Figs. 7 and 8 show PD performances as a function of SNR. Theoretical and simulated detection probabilities of the proposed method are shown in Fig. 7. In addition, there are simulated PD results of the Tufvessons method. It can be seen that simulated and theoretical results of the proposed scheme coincide very well. The performances of the proposed method and the Tufvessons method are almost the same too. The PD performances of the proposed method and the S&C method are compared in Fig. 8. It can be seen that the S&C method has quite large performance loss ( 10 dB). Despite of a broad timing metric of the S&C method (Fig. 5), it achieves probability 1 because PD just tells the probability that the threshold is exceeded when a sample is taken from the correct time instant. However, the S&C method tends to raise false alarm rate due to broad timing metric. Figs. 9 and 10 show PM ax and PD,M ax performances as a function of SNR. Theoretical and simulated results of the proposed method are shown in Fig. 9. Simulated results of the

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC07)

1 Nc = 256 0.9 L = 128 0.8 TCP = 0 0.7 0.6 PD 0.5 0.4 0.3 0.2 0.1 0 30 f = 0

1 Proposed method, 101 0.9 0.8 0.7 Theory, 10 Simul., 10

1 1 1

S&C 101 Proposed method, 102 S&C 10 S&C 10 S&C 10

2

Proposed method, 103

3

0.6 PD 0.5 0.4

Proposed method, 104

4

Simul., Tufv. 10 Theory, 102 Simul., 102 Simul., Tufv. 10 Theory, 10 Simul., 10
3 3

f = 0 0.3 TCP = 0 0.2 L = 128 0.1

3 Simul., Tufv. 10

Theory, 10

Simul., 104 Simul., Tufv. 10 25 20 15 SNR [dB] 10 5

4

Nc = 256 0 0 30 25 20 15 SNR [dB] 10 5 0

Figure 7: PD results with different PF A values. Tufvessons method are shown also. It can be seen that simulated and theoretical results of the proposed scheme coincide very well again. The performances of the proposed method and the Tufvessons method are almost the same too. PM ax and PD,M ax performances of the proposed method and the S&C method are compared in Fig. 10. It is observed that now there is a big difference in the performance. The S&C method does not attain probability 1 because a maximum sample may be selected easily from the incorrect time instant. Frequency synchronization performance of the proposed method is evaluated in Fig. 11, where f is 0.1 subcarrier spacings. It can be seen that the proposed method achieves theoretical CRB when Ec /N0 > 10 dB.

Figure 8: PD results with different PF A values. R EFERENCES

[1] T. M. Schmidl and D. C. Cox, Robust Frequency and Timing Synchronization for OFDM, IEEE Transactions on Communications, vol. 45, no. 12, pp. 1613 - 1621, Dec. 1997. [2] M. Speth, S. A. Fechtel, G. Fock and H. Meyr, Optimum Receiver Design for Wireless Broad-Band Systems Using OFDM Part - I, IEEE Transactions on Wireless Communications, vol. 2, no. 3, pp. 424 - 430, May 2003. [3] M. Sandell, J. J. van de Beek and P. O. B orjesson, Timing and Frequency Synchronization in OFDM Systems Using the Cyclic Prex, in Proceedings of the IEEE International Symposium on Synchronization, 1995, pp. 16-19. [4] J. J. van de Beek, M. Sandell and P. O. B orjesson, ML Estimation of Time and Frequency Offset in OFDM Systems, IEEE Transactions on Signal Processing, vol. 45, no. 7, pp. 1800 - 1805, July 1997. [5] H. Minn, M. Zeng and V. K. Bhargava, On Timing Offset Estimation for OFDM Systems, IEEE Communication Letters, vol. 4, no. 7, pp. 242 - 244, July 2000. [6] B. Park, H. Cheon, C. Kang and D. Hong, A Novel Timing Estimation Method for OFDM Systems, IEEE Communication Letters, vol. 7, no. 5, pp. 239 - 241, May 2003. [7] J. L. Zhang, M. Z. Wang and W. L. Zhu, A Novel OFDM Frame Synchronization Scheme, in Proceedings of the IEEE International Conference on Communications, Circuits and Systems and West Sino Expositions, 2002, pp. 119-123. [8] F. Tufvesson, O. Edfors and M. Faulkner, Time and Frequency Synchronization for OFDM Using PN-sequence Preambles, in Proceedings of the IEEE Vehicular Technology Conference, 1999, vol. 4, pp. 2203-2207. [9] J. E. Kleider and S. Gifford, Synchronization for Broadband OFDM Mobile Ad Hoc Networking: Simulation and Implementation, in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, 2002, vol. 4, pp. 3756-3759. [10] J. G. Proakis, Digital Communications. McGraw-Hill, Inc., New York, USA, 1995. [11] J. Iinatti, On the Threshold Setting Principles in Code Acquisition of DS/SS Signals IEEE Journal on Selected Areas in Communications. vol. 18, no. 1, pp. 62-72, Jan. 2000.

V.

C ONCLUSIONS

A Matched lter based time and frequency synchronization method for OFDM systems was proposed, which provides fast synchronization using PN-sequence preambles. The proposed method transmits a known PN-preamble in the frequency domain and it has two identical halves in the time domain. Because MF in the receiver is matched to one half of this time domain preamble, two identical autocorrelation functions are achieved from the MF output. These two autocorrelation functions are used for timing and frequency synchronization. In timing synchronization, these autocorrelation functions are combined noncoherently. In frequency synchronization, a phase difference between those is estimated. The obtained results indicate that the proposed method attains the optimum theoretical performance.

The 18th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC07)

1 Nc = 256 0.9 L = 128 0.8 TCP = 0 0.7 f = 0

D,Max PMax, theory PMax, simul. PMax, simul., Tufv. PD,Max, theory, 101 PD,Max, simul., 101
1 PD,Max, simul., Tufv., 10

0.6 0.5 0.4 0.3 0.2 0.1 0 30

&P

Max

PD,Max, theory, 10

PD,Max, simul., 102 PD,Max, simul., Tufv., 10 PD,Max, theory, 103 P

D,Max 2

, simul., 103

3 PD,Max, simul., Tufv., 10

D,Max

, theory, 104

PD,Max, simul., 104 P

D,Max 4 , simul., Tufv., 10

25

20

15 SNR [dB]

10

10

Simulated CRB 10 Frequency error variance

1

Figure 9: PM ax and PD,M ax results with different PF A values.

10

10

10

1
P
Max

, Proposed method
1

0.9 0.8 0.7

D,Max

PMax, S&C PD,Max, Proposed method, 10 P P P P

D,Max D,Max D,Max D,Max

10 20 Nc = 256 L = 128

15

10

5 SNR [dB]

10

, S&C, 101 , Proposed method, 102 , S&C, 102 , Proposed method, 103

Figure 11: Performance of the frequency offset estimator.

TCP = 0 f = 0
4

0.6 0.5 0.4 0.3 0.2 0.1 0 30

PD,Max, S&C, 103 PD,Max, Proposed method, 10 PD,Max, S&C, 10

4

Max

&P

25

20

15 SNR [dB]

10

Figure 10: PM ax and PD,M ax results with different PF A values.