580

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 46, NO. 5, MAY 1998

On the Effect of Wiener Phase Noise in OFDM Systems

Luciano Tomba, Member, IEEE

Abstract Multicarrier modulation exhibits a signicant sensitivity to the phase noise of the oscillator used for frequency down-conversion at the portable receiver. For this reason, it is important to evaluate the impact of the phase noise on the system performance. In this letter we present an accurate method to determine the error probability of an orthogonal frequencydivision multiplexing (OFDM) system in the presence of phase noise. In particular, four modulation schemes are analyzed and their performances are compared. Index Terms OFDM, phase noise.

I. INTRODUCTION

RTHOGONAL frequency-division multiplexing (OFDM) has been suggested in a recent paper [1] for the transmission of asynchronous transfer mode (ATM) data packets over indoor radio channels. The main advantage of OFDM is to split the available signal bandwidth into a large number of independent narrow-band subchannels. By selecting the subchannel bandwidth smaller than the coherence bandwidth of the frequency-selective channel, each subchannel experiences at fading [2]. This work limits itself to consider the effect of Wiener phase noise in OFDM systems. Some analysis of the effect of phase noise in OFDM systems has been presented in [3]. However, in [3] the authors do not derive exact error probability formulas because they approximate the contribution of the phase noise with an additional noise component. Their approach is accurate enough in cases where the linewidthsymbol period product is sufciently small (and the signal-to-noise ratio large) such that the error rate degradation is mainly determined by the intercarrier interference. Instead, the analysis presented in this letter takes into account the exact statistic of the phase noise to evaluate the error probability. Lastly, the analysis is used to compare the performance of four modulation schemes [binary phase-shift keying (BPSK), quarternary phase-shift keying (QPSK), differential BPSK (DBPSK), and differential QPSK (DQPSK)]. II. SYSTEM MODEL

overlap spectrally, but since they are orthogonal within a , symbol duration (the th carrier frequency is is some reference frequency and ) where the signal associated with each sinusoid can be recovered as long as the channel does not destroy their orthogonality. In practice, the samples of the OFDM signal are generated by taking the inverse discrete Fourier transform (IDFT) of a discrete-time input sequence and passing the transform samples through a pulse-shaping lter; at the receiver side, dual transformations are implemented. By assuming an additive white Gaussian noise (AWGN) channel, after frequency down-conversion, the received signal is given by [3], [4] (1) ( is the imaginary unit), where is the additive noise, and is the multiplicative error term due to the oscillator phase noise.1 As it is usual in theoretical papers on the subject (e.g., [3] and [5]), to account for the phase noise is assumed to be a continuous-path Brownian motion (or WienerL evy process) with zero mean (the parameter represents the two-sided and variance 3-dB linewidth of Lorentzian power density spectrum of the oscillator). The above mathematical model is accurate enough for modeling the phase noise process and it is suitable to get a quantitative measure of the system performance degradation due to the phase noise by analytical methods. However, it must be noted that this approach is valid as long as the baseband conversion of the OFDM signal is performed by a freerunning oscillator.2 Instead, when the baseband conversion is performed by means of an oscillator which is properly locked to the carrier of the received signal, the variance of the phase error does not increase without limit and the numerical results presented in this letter do not apply. is multiplied by tones In the receive unit the signal with the same frequency as in the transmitter and sent to the integrate-and-dump (I&D) lters with integration input of time . In conclusion, the signal at the output of the th I&D lter is [3] (2)

Basically, an OFDM signal [2], [4] consists of the superposinusoidal subcarriers with frequency spacing ; sition of each subcarrier is modulated by symbols with period equal to the inverse frequency spacing. The modulated subcarriers
Paper approved by O. Andrisano, the Editor for Modulation for Fading Channels of the IEEE Communications Society. Manuscript received February 13, 1997; revised June 15, 1997 and January 15, 1998. This work was supported in part by the Italian Research Ministry (MURST) and by the Consiglio Nazionale delle Ricerche (CNR), Rome, Italy. The author is with the Dipartimento di Elettronica e Informatica, Universit` a di Padova, 35131 Padova, Italy (e-mail: luciano@dei.unipd.it). Publisher Item Identier S 0090-6778(98)03855-0.

1 In principle, the phase noise of both of the oscillators (respectively, at the base station and at the portable terminal) should be accounted for. However, in practice, the oscillator used at the base station is sufciently stable to disregard its phase noise. 2 The numerical results presented in this letter (and the formulas of Appendixes A and B) have been obtained by assuming the initial condition #(0) = 0.

00906778/98$10.00 1998 IEEE

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 46, NO. 5, MAY 1998

581

where (3) and are samples of an additive Gaussian noise. In (2) the is multiplied by which depends information quantity on the phase noise, while it is independent of the particular subcarrier index. The summation in (2), instead, represents the . Note interbin interference [3] which is zero when that this interference depends on data from all of the different subchannels. III. PERFORMANCE EVALUATION To compare different modulation schemes, the performance measure considered in this work is the probability of bit error (BER). In particular we consider BPSK and QPSK, and their differential versions (DBPSK and DQPSK, respectively). All symbols are assumed independent and identically distributed. The following analysis makes use of some basic results presented in [5] and in [6] which are reported, for convenience sake, in the appendixes. BPSK: In this case the decision variable for the th subcarrier is given by

The saddlepoint approximation [7] is a fast and accurate procedure to calculate integrals like (6) and it is used in this work. QPSK: The decision variable is (9) To get the BER it is helpful to use the union bound [8]; hence, we get (10) denotes the decision region for the symbol where Once again, use of the MGF yields .

(11) where (12) The analytical expression of while is the same as . Finally, the BER is with is given by (17), replaced by

(4) The statistic of is given in Appendix A. Nevertheless, it is not easy to nd the statistic of . For this reason, we approximate by a complex Gaussian random variable (RV) whose mean and variance are evaluated in Appendix B. By using the Smirnov test, by simulation it can be veried that the signicance levels are, approximately, between 0.70.9. In is a zero-mean Gaussian RV with conclusion, variance and the BER is3

(13) DBPSK: One possible implementation of this modulation scheme makes use of the samples with the same subcarrier index in the previous discrete Fourier transform (DFT) block as a phase reference. In the presence of phase noise this method does not represent a good solution because the phase rotation on the information data is not the same in different DFT blocks. To overcome this problem, another technique consists in performing detection of one symbol transmitted on a given subcarrier using an adjacent subcarrier in the same DFT block as [2]. Let be the differentially encoded data from the source ; the decision variable is ( stands for complex bits and conjugate) (14) From (14) it is apparent that the phase noise does not produce any phase rotation on the product of the information symbols and , but it modies only its magnitude. The BER is

(5) Each term between curved brackets of (5) can be evaluated by using the moment generating function (MGF) of the RV involved

(6) In (6) namely is the MGF of conditional to ,

(7) where

(15) involves RVs which are assumed The decision variable independent of each other. The error probability conditional to a pair and is known [8]. The average with is simply obtained by using the method of the respect to Gaussian quadrature rules because the moments of can be calculated using the MGF reported in Appendix A.

(8)
and =[1], as well as the subscripts R and I , stand for real and imaginary part, respectively.
3 <[1]

582

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 46, NO. 5, MAY 1998

Fig. 1. BER of BPSK for the AWGN channel in the presence of phase noise for different values of .

Fig. 3. BER of DBPSK for the AWGN channel in the presence of phase noise for different values of .

Fig. 2. BER of QPSK for the AWGN channel in the presence of phase noise for different values of .

Fig. 4. BER of DQPSK for the AWGN channel in the presence of phase noise for different values of .

DQPSK: This modulation format differs from the above DBPSK only in the size of the constellation. The formulas to compute the error probability are a straightforward extension of the one for QPSK and DBPSK; hence, they are omitted.

IV. COMMENTS

ON THE

RESULTS

AND

CONCLUSION

For each modulation scheme, the BER is shown in Figs. 14. Different curves correspond to different values of which depends on the oscillator linewidth, on the number of subcarriers, and on the symbol period. We have assumed ideal removal of any phase and frequency offset at the receiver side; hence, the transmitted symbol is impaired only by the common phase noise, i.e., the equal multiplicative term on each subcarrier.

It is apparent that the performance strongly degrades for in the case of BPSK. As expected, even worse results are obtained for QPSK. Fig. 3 shows the BER for DBPSK. Although this modulation scheme is less efcient than BPSK, it is robust enough against the common phase noise when compared to BPSK (see Fig. 1). The performance degradation is mainly due to the additional amount of additive noise corresponding to the interbin interference . The performance of DQPSK is better than QPSK but worse than DBPSK; hence, DBPSK seems to be the most robust scheme against the phase noise. In conclusion, the methodology presented in this work is general and accurate. Moreover, with proper changes, it can be applied to different modulation schemes also in the presence of multipath fading channels (by using a semianalytical approach).

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 46, NO. 5, MAY 1998

583

APPENDIX A MGF OF Let (16) When [5], the exponential function in (16) can be . The joint expanded in a power series as MGF of the real and imaginary parts of (16) is given by [5]

Moreover, by extending the results presented in [6], we get (20) is the inverse Laplace transform of the where argument evaluated at . ACKNOWLEDGMENT The author would like to thank the anonymous reviewers for their careful revision and important suggestions which signicantly helped to improve the presentation of this letter. REFERENCES
[1] A. Chini, M. S. El-Tanany, and S. A. Mahmoud, Transmission of high rate ATM packets over indoor radio channels, IEEE J. Select. Areas Commun., vol. 14, pp. 460468, Apr. 1996. [2] H. Sari, G. Karam, and I. Jeanclaude, Transmission techniques for digital terrestrial TV broadcasting, IEEE Commun. Mag., vol. 33, pp. 100109, Feb. 1995. [3] T. Pollet, M. Van Bladel, and M. Moeneclaey, BER sensitivity of OFDM systems to carrier frequency offset and Wiener phase noise, IEEE Trans. Commun., vol. 43, pp. 191193, Feb./Mar./Apr. 1995. [4] J. A. C. Bingham, Multicarrier modulation for data transmission: An idea whose time has come, IEEE Commun. Mag., vol. 28, pp. 514, May 1990. [5] G. J. Foschini and G. Vannucci, Characterizing ltered light waves corrupted by phase noise, IEEE Trans. Inform. Theory, vol. 34, pp. 14371448, Nov. 1988. [6] G. L. Pierobon and L. Tomba, Moment characterization of phase noise in coherent optical systems, J. Lightwave Technol., vol. 9, pp. 9961005, Aug. 1991. [7] C. W. Helstrom, Approximate evaluation of detection probabilities in radar and optical communications, IEEE Trans. Aerosp. Electron. Syst., vol. AES-14, pp. 630640, July 1978. [8] J. G. Proakis, Digital Communications, 2nd ed. New York: McGrawHill, 1989.

(17) where , , and APPENDIX B AND VARIANCE OF , . .

MEAN

Because of the assumptions on the data Hence

(18)

(19)