IEEE TRANSACTIONS ON BROADCASTING, VOL. 46, NO. 3, SEPTEMBER 2000 215 OFDM System Channel Estimation Using Time-Domain Training Sequence for Mobile Reception of Digital Terrestrial Broadcasting Che-Shen Yeh, Yinyi Lin, and Yiyan Wu, Member, IEEE Abstract In this paper, a channel estimation technique using time-domain training sequence is investigated. It is demonstrated that this technique can estimate the channel status more accurately, under low C/N situations, than the conventional frequency domain technique using in-band pilots [1], [2]. A novel channel estimation and compensation scheme for fast fading channel is developed for OFDM system using time domain training sequence. It is compared with the frequency domain pilot channel estimation scheme. Index Terms Channel estimation, COFDM, data broadcasting, digital television, mobile reception, TCM, terrestrial broadcasting. I. INTRODUCTION THERE has been ever increasing interests in the mobile reception of digital television and data broadcasting signals. In the mobile reception environment, the field strength of a received signal varies with respect to time, or space. When the operating frequency becomes higher, the fading becomes more severe. The received signal can be artificially separated into two parts: a long-term fading, and a short-term fading, i.e., The long-term fading is mainly caused by terrain configuration, and the building environment surrounding the transmitter and the mobile unit. Terrain configurations can be classified as open area, flat terrain, hilly terrain and mountain area. The man-made environment can be classified as rural area, quasisuburban area, suburban area and urban area. Short-term fading is mainly caused by multipath reflections of a transmitted signal by local scatters such as buildings, and other man-made structures, or by natural obstacles surrounding a mobile unit [3]. This paper investigates different OFDM channel estimation schemes for mobile reception. These channel estimation methods are [1]: A) Frequency domain Pilot Frequency Interpolation (FPFI); Manuscript received July 28, 2000; revised September 12, 2000. C.-S. Yeh and Y. Lin are with the Department of Electrical Engineering National Central University, Chung Li Taiwan 32054, R.O.C. Y. Wu is with the Communications Research Centre Canada Ottawa, Ontario, Canada, K2H 8S4. Publisher Item Identifier S 0018-9316(00)11640-3. (1) B) Time domain Pilot Time Cross-correlation (TPTC), C) Time domain Pilot Time Cross-correlation and Time varying Estimation (TPTCTE). For a slow fading channel (assuming channel model does not change over one OFDM symbol duration), the TPTC method performs better than the FPFI [1] under low C/N situations. The focus of this paper is to improve TPTC method over fast fading channel, which results in the proposal of TPTCTE scheme. Section II discusses different channel model of mobile reception. Section III describes different OFDM channel estimation methods. Section IV provides the analysis of simulation results. Section V is the conclusion. II. MOBILE RECEPTION CHANNEL OF DIGITAL BROADCASTING IN VHF/UHF BANDS As mentioned before, the received signal can be modeled as a long-term fading, and a short-term fading. It has been approved that a long-term fading can be modeled as a lognormal distribution [3], [4], assuming there is no direct signal path, but reflected signals, coming form arbitrary directions with equal probability. The signal received at the mobile receiver becomes where is the transmitting frequency, is the velocity of the mobile unit, is the direction of th wave arrival, is called the wave number, and is the wavelength, and is the delay length of th wave arrival. Expression is the complex exponential representing a transmitting frequency that propagates in time domain. represents a fading signal, that is a complex random variable with zero mean and variance of one. is also a random variable ranging between 0 to 360. Equation (2) represents a fading signal and the maximum fading frequency can be obtained as [3], [4] (2) (3) 0018 9316/00$10.00 2000 IEEE
216 IEEE TRANSACTIONS ON BROADCASTING, VOL. 46, NO. 3, SEPTEMBER 2000 Fig. 3. Channel estimation of time pilot in the time domain. Fig. 1. Channel 2 mobile fading path with V =100km/hr; f = 600 MHz. Fig. 2. Channel model estimation of frequency pilot in the frequency domain. Level-crossing rate (lcr) is defined as the number of crossing at the positive slops at a level and the total number of crossing over a time duration, i.e., the lcr is where is the envelope of the signal strength with respect to its rms value. The average duration of fade (adf),, is defined as the sum of fades at level divided by the total number of crossings [3]. The effects of long-term fading and short-term fading are shown in Fig. 1 [ km/hr, MHz,, adf( 20 db) 0.0072, lcr( 20 db) 7.42]. The relation of lcr and, from the two channel models used in this paper (see later Section) and the four mobile speed and carrier frequency conditions ( km/hr or 100 km/hr, MHz or 600 MHz), will generate different fading paths, and can be obtained from the above equation as 5.5, 13.9, 22.2, and 55.6 Hz, respectively. III. CHANNEL ESTIMATION METHODS OF OFDM SYSTEM DESCRIPTION A typical OFDM system is presented in Fig. 2. The receiver estimates the channel model, and uses it to equalize the received data symbol. In this section, three channel estimation schemes are discussed for the fast-fading channels. A. Method I (FPFI) In the transmitter, a multiplexer (MUX) combines the data symbols and pilot signals into a data stream in the frequency (4) domain, and the data stream is fed into an Inverse Fast Fourier Transform (IFFT) block. The IFFT block converts the signal from frequency domain to the time domain. The guard interval is, then, inserted before each OFDM symbol as the circular extension of the signal itself, to eliminate the Inter-Symbol Interference (ISI) between the OFDM symbols. Scattered pilot tones are BPSK modulated with a PN sequence in the frequency domain and are inserted into each OFDM symbol with a ratio of. The channel response is estimated in the frequency domain by comparing the received pilots with the locally stored reference scattered pilots. This method is referred as the Frequency Pilot Frequency Interpolation (FPFI) technique, since the channel response is obtained via frequency interpolation [1], [2]. B. Method II (TPTC) In the second technique, a PN sequence is inserted at the beginning of an OFDM symbol in the time domain. This technique is referred as the Time Pilot Time Correlation (TPTC) technique as described in Fig. 3. The TPTC multiplexes the data symbols and PN sequences in the time domain, with each OFDM symbol having a PN sequence. The guard interval and the training sequence are inserted in front of the data sequence to form a TPTC-OFDM symbol. In the receiving end, the channel response is estimated using the cross correlation of received PN sequence and the locally stored PN sequence. If the transmitted PN sequence is, the received pilot signal can be expressed as in the time domain (where represents the convolution operator) or in the frequency domain. Thus the channel response is given by. As the spectrum of a PN sequence is nearly constant in the frequency domain, can be approximated as a constant,, which is proportional to the length of the PN sequence [5]. C. Method III (TPTCTE) In mobile reception, the amplitude and phase of a channel response could vary within the interval of one OFDM symbol, while the signal amplitude maintains a log-normal distribution, and the phase is varying with a rate, which is dependent on the incident angle and Doppler shift of the echo. The Doppler shift is dependent on the frequency of carrier and the speed of mobile unit, as expressed in Eq. (2). In mobile reception, significant Doppler shift will result in serious estimation and equalization errors. In order to improve the performance of the TPTC method, a new method called TPTCTE (Time Domain Pilot
YEH et al.: OFDM SYSTEM CHANNEL ESTIMATION USING TIME-DOMAIN TRAINING SEQUENCE FOR MOBILE RECEPTION 217 TABLE I OFDM SYSTEM PERFORMANCE SIMULATION PARAMETERS OF MOBILE RECEPTION Fig. 4. Proposed channel estimation technique (TPTCTE). Time Domain Cross correlation Time Varying Estimation) is proposed as shown in Fig. 4. Comparing with Fig. 3, a function block is added to estimate and compensate the effect of short-term fading. Both the amplitude and the phase change due to Doppler shift are estimated. In this study, the time duration of an OFDM symbol is divided into equally spaced sub-symbol time slots ( or 32 in our study). Assuming the echo delay does not change in one OFDM symbol duration, and the change of amplitude and phase is negligible during a sub-symbol time slot. The TPTCTE method is described as: 1) In the receiving end, the estimated channel response is obtained by cross-correlation of the received training signal and the locally stored PN sequence. Offset the DC bias of the cross-correlation process. The estimated channel response is expressed as where and is the estimated amplitude and phase of the th echo at the beginning of the th OFDM symbol. 2) The echo which has the maximum amplitude is referred as the estimated main signal at the beginning of the th OFDM symbol,. 3) Repeating Step 1 and 2 for the training sequence that follows the present OFDM symbol to obtain and. 4) Calculating the long term fading estimation factor. Since the long term fading is modeled as log-normal distribution,, where is the number of sub-symbols in an OFDM symbol duration. The estimated signal strength of the th sub-symbol will be for. 5) Compensating long term fading in time-domain by dividing each data sample in one of the sub-symbol duration with corresponding. 6) Normalizing and to the corresponding main signals (making the largest signal/echo amplitude equals to 1) to form and. 7) Assuming the echo delay are relatively unchanged in the same OFDM symbol, forming channel responses, where ; based on and, via linear interpolation of echoe s (5) amplitudes and phases. The amplitude of the th echo in the th sub-symbol duration is estimated as (6) where, and, are echo amplitudes that can be obtained from estimated channel responses and. Similarly, the estimated phases of the th echo in the th sub-symbol duration will be (7) where and are echo phases from and. 8) From, calculating channel frequency responses, each of which corresponds to one of sub-symbol duration. 9) Averaging the frequency responses,, to obtain one estimated channel frequency response. 10) Using to equalize the received OFDM data in frequency domain. IV. PERFORMANCE COMPARISONS AND DISCUSSION The channel models used in the simulation are two fading channels with different levels of Additive White Gaussian Noise (AWGN) injection. A. Mobile Reception without TCM Two mobile speeds (40 km/hr and 100 km/hr) and two carrier frequencies (150 MHz and 600 MHz) are assumed in the simulation. There are four combinations of the Doppler shift, i.e., 5.5, 13.9, 22.2 and 55.6 Hz. The OFDM system parameters used for simulations are summarized in Table I. All data carriers in the OFDM are 16 QAM modulated signals [signal constellation ( ) with ], and the average power of a data symbol is. The pilot signals with amplitude 4.5 are used, or 3 db higher pilot signal power than the average data signal power. For the short-term fading channel, multipath channel model Ensemble F in the US Grand Alliance DTV tests [6] is used, i.e., (8)
218 IEEE TRANSACTIONS ON BROADCASTING, VOL. 46, NO. 3, SEPTEMBER 2000 Fig. 5. SER vs. SNR for mobile Channel 1 [mobile condition: f =5:5 Hz, V =40km/hr, f = 150 MHz, lcr(010 db) = 0.293, adf(010 db) = 0.0191]. (a) where the unit delay seconds. In addition, the simplified nonhilly urban channel model recommended for DAB test (pr EN50248) is also used in simulations as channel Model 2 [7]: (9) When the channel model changes slowly, the performamnces of TPTC and TPTCTE are almost identical and they all outperform FPFI method, as shown in Fig. 5, due to the integration operation conducted in time-domain cross-corelation process, which can reduce the impact of the AWGN [1]. Similar results were obtained for the channel Model 2. When the carrier frequency is high, and lcr are high. The channel response, during an OFDM symbol period, is time varying. The channel model estimated using in-band pilots might not represent what data carriers experienced. These will result in irreducible error floor, see Fig. 6(a) and (b). When and lcr are high, the channel responses estimated using in-band pilots and real channel model differ in the same OFDM symbol, which is slightly different from the TPTC assumption. The precision of the TPTC is obviously lower than the TPTCTE. When S/N is higher, the FPFI performs better than the TPTC, because the PN sequence of the TPTC is at the beginning of the OFDM symbol. This will result in more errors when channel response is time varying during one OFDM symbol period. In low S/N condition, the cross correlation process in the TPTC scheme can reduce the impact of the AWGN, while the in-band pilots of the FPFI become contaminated by the AWGN. The TPTC can out perform FPFI in the low S/N region [see Fig. 6(a) and (b)]. Channel response variation is considered by TPTCTE, which results in more precise channel estimation. The TPTCTE has better performance in both slow-fading and fast-fading environments. (b) Fig. 6. (a) SER vs. SNR for mobile Channel 1 [mobile condition: f = 55:5 Hz, V = 100 km/hr, f = 600 MHz, lcr(015 db) = 0.1953, adf (015 db) = 0.0004]. (b) SER vs. SNR for mobile Channel 2 [mobile condition: f =22:2 Hz, V =40km/hr, f = 600 MHz, lcr(020 db) = 3.9062, adf(020 db) = 0.0145]. From the Fig. 6(a) and (b), the channel Model 2 ( ) is smaller in Fig. 6(b). But, the lcr and adf values are larger than. The reason is that the echo amplitudes are high in, that the echo time-variant is larger with the same Doppler shift or, which leads to more severe fading. Therefore, the with smaller Doppler shift and the with larger Doppler shift have similar performances. B. Mobile Reception with TCM In order to improve the mobile reception performance, trellis coded modulation (TCM) and interleaving are usually implemented with an OFDM modulation system. The resulted system diagram is shown in Fig. 7, where a pair of 4-state one-dimensional trellis encoders [8], see Fig. 8, is used for in-phase and
YEH et al.: OFDM SYSTEM CHANNEL ESTIMATION USING TIME-DOMAIN TRAINING SEQUENCE FOR MOBILE RECEPTION 219 Fig. 7. Block diagram of OFDM system with 2/3 TCM coding. (a) Fig. 8. OFDM trellis encoder and symbol mapped. (a) Fig. 10. (b) (a) Mobile Channel 1 with 2/3 TCM SER v.s. SNR [mobile condition: f =22:2, V =40 km/hr, f = 600 MHz, lcr(015 db) = 0.1953, adf(015 db) = 0.0011]. (b) Mobile Channel 2 with 2/3 TCM SER v.s. SNR [mobile condition: f = 22:2, V = 40 km/hr, f = 600 MHz, lcr(020 db) = 3.125, adf(020 db) = 0.0124]. Fig. 9. (b) (a) SER vs. SNR for mobile Channel 1 with 2/3 TCM [mobile condition: f = 55:5 Hz, V = 100 km/hr, f = 600 MHz, lcr(015 db) = 0.3906, adf(015 db) = 0.0005]. (b) SER vs. SNR for mobile Channel 2 with 2/3 TCM [mobile condition: f = 22:2 Hz, V = 40 km/hr, f = 600 MHz, lcr(020 db) = 3.125, adf(20 db) = 0.0062]. quadrature components of the QAM-COFDM signal, respectively. A system operator may choose various interleaving depth depending on the characteristics of the channel and services. A longer interleaving depth is preferred for a larger average duration of fade and for channels with strong impulse noise interference. In this study, two values of interleaving depth 16 and 256 are considered for simulations. In Fig. 7, after the 2/3 TCM encoder the real and imaginary data parts become two sets of 3 bits/symbol streams to form a 64 QAM symbol stream. After a block interleaving (depth of 16 or 256), the signals are fed to an OFDM modulator. From Fig. 9(a) and (b), due to large estimation errors for fast fading channel, TPTC scheme results in a transmission error floor, even with TCM. By using TPTCTE channel estimation with TCM coding, the error rate is greatly reduced and the error floor is eliminated. The adf represents duration of the signal fade. If adf value is high, it indicates more long bursts of errors, which cause irreducible error floor. Deeper interleave might be needed to mitigate the long burst of errors. Fig. 10(a) and (b) present examples with slower receiving terminal speed, which means longer adf. Two interleavers depths were used: 16 and
220 IEEE TRANSACTIONS ON BROADCASTING, VOL. 46, NO. 3, SEPTEMBER 2000 256. It can be seen that the larger interleaver can further improve the performance of the COFDM system in a fading channel. V. CONCLUSION Channel estimation using time domain training sequence (TPTC) performs better than the frequency domain pilot method (FPFI) in low S/N cases. TPTC method assumes that channel response does not change during an OFDM symbol period. This is acceptable for slow fading channel. The channel estimation errors of TPTC method may be severe in cases of fast fading. TPTCTE method compensates channel response based on sub-blocks of the FFT symbol. The simulation results show that the TPTCTE method is able to improve the performance for time varying channel, and reduces SER error floor. The TPTCTE has better performance in both slow-fading and fast-fading environments. Larger interleaving depth can improve the performance for mobile reception. Better coding and interleaving system, such as parallel concatenated trellis code, might further improve the performance. [11] H. Sari, G. Karam, and I. Jeanclaude, Frequency domain equalization of mobile radio and terrestrial broadcast channels, in IEEE Global Telecommunication Conference, vol. 1, 1994, pp. 1 5. [12] Y. Wu and W. Y. Zou, COFDM: An overview, IEEE Transactions on Broadcasting, vol. 41, no. 1, pp. 1 7, March 1995. [13] R. Negi and J. Cioffi, Pilot tone selection for channel estimation in a mobile OFDM system, IEEE Transactions on Consumer Electronics, vol. 44, no. 3, pp. 1122 1128, August 1998. [14] M.-S. Kang and W.-J. Song, A robust channel equalizer for OFDM TV receivers, IEEE Transactions on Consumer Electronics, vol. 44, no. 3, pp. 1129 1133, August 1998. Che-Shen Yeh received the B.S. and M.S.E.E. in electrical engineering from Chung-Yuan Christian University, Jung-Li, Taiwan in 1980 and 1986 respectively. He is currently a Ph.D. candidate in electrical engineering at the National Central University, Jung-Li, Taiwan. His major research interests are digital transmission for television and audio. He leads the DTV/DAB transmission project at the Telecommunications Laboratories of Chunghwa Telecom Co., Ltd. Taiwan, and was also responsible for DTV field trials and the development of DTV transmission system standards. He was a member of the Task Force for SDTV/HDTV transmission standards and regulation at the Ministry of Transportation and Communications, Taiwan. REFERENCES [1] C. S. Yeh and Y. Lin, Channel estimation using pilot tons in OFDM systems, IEEE Trans. on Broadcasting, vol. 45, no. 4, pp. 400 409, Dec. 1999. [2] Digital video broadcasting (DVB) framing structure, channel coding and modulation digital terrestrial television (DVB-T),, ETS 300 744, v1.2.1, June 1999. [3] W. C.Y. Lee, Mobile Communications Design Fundamental. New York, NY: John Wiley & Sons, 1993. [4], Mobile Communications Engineering. New York, NY: Mc- Graw-Hill, 1982. [5] Y. Lin, Shift and added property of m-sequence and its application to channel characterization of digital magnetic recording, IEE Proceedings Communications, vol. 142, no. 3, pp. 135 140, June 1995. [6] Grand alliance transmission system selection test plan,, SSWP2-1218, Rev. 13, January 1994. [7] Characteristics of DAB receiver, European Committee for Electro- Technical Standards, PREN50248, 1996. [8] Y. Wu and W. Y. Zou, Orthogonal frequency division multiplexing: A multiple-carrier modulation scheme, IEEE Transactions on Consumer Electronics, vol. 41, no. 3, August 1995. [9] A. Chini, Y. Wu, M. El-Tanany, and S. Mahmoud, An OFDM-based digital ATV terrestrial broadcasting system with a filtered decision feedback channel estimator, IEEE Transactions on Broadcasting, vol. 44, no. 1, pp. 2 11, March 1998. [10] G. Ungerboeck, Trellis coded modulation with redundant signal sets, IEEE Communications Mag., vol. 27, pp. 5 21, Feb. 1987. Yinyi Lin was born in Chuang Hua, Taiwan, R.O.C., in 1959. He received the B.S. degree in communication engineering from National Chiao Tung University, Hsin Chu, Taiwan in 1981, and the M.S. and Ph.D. degrees in electrical engineering from the University of California, San Diego in 1986 and 1989, respectively. During 1988, he spent six months at the IBM Almaden Research Center, San Jose, CA working in the area of signal processing for digital magnetic recording. From 1990 to 1991, he was with Archive Corporation, Costa Mesa, CA, where he engaged in the research of quarter inch cartridge (QIC) tape system. Since 1991, he has been with the Electrical Engineering Department of National Central University, Jung-Li, Taiwan, where he is now a Professor. His current research interests include coding and signal processing for digital magnetic recording systems, digital compression, and error correction for documents. Yiyan Wu received the M. Eng. and Ph. D. degrees from Carleton University, Ottawa, Ontario, Canada in 1986 and 1990 respectively. He is a Senior Research Scientist with the Communications Research Centre, Ottawa, Canada. His research interests include digital video compression and transmission, high definition television (HDTV), signal and image processing, satellite and mobile communications. He is actively involved in the ATSC technical and standard activities and ITU-R digital television and data broadcasting studies. He is an Adjunct Professor at Carleton University, Ottawa, Canada, and a Member of the Administrative Committee of the IEEE Broadcast Technology Society (BTS). He also represents the IEEE BTS to the ATSC Executive Committee.