5, MAY 1998


Effects of Clipping and Filtering on the Performance of OFDM
Xiaodong Li and Leonard J. Cimini, Jr., Senior Member, IEEE
Abstract—Orthogonal frequency division multiplexing (OFDM) is an attractive technique for wireless communication applications. However, an OFDM signal has a large peak-to-mean envelope power ratio, which can result in significant distortion when passed through a nonlinear device, such as a transmitter power amplifier. We investigate, through extensive computer simulations, the effects of clipping and filtering on the performance of OFDM, including the power spectral density, the crest factor, and the bit-error rate. Our results show that clipping and filtering is a promising technique for the transmission of OFDM signals using realistic linear amplifiers. Index Terms— Clipping, crest factor, filtering, orthogonal frequency division multiplexing (OFDM).

II. SYSTEM DESCRIPTION In our simulation, we assume an OFDM signal with subchannels and a time guard interval of length 32. Quadrature phase shift-keying (QPSK) data symbols are passed through an inverse fast Fourier transform (IFFT) module to realize the OFDM modulation. If the digital OFDM signals are clipped directly, the resulting clipping noise will all fall in-band and cannot be reduced by filtering. To address this aliasing problem, in this simulation, we oversample the OFDM signal by a factor of 8. In addition, to reduce the implementation complexity, the complex-valued baseband OFDM signal is modulated up to a carrier frequency equal to 1/4 of the sampling frequency. Then, the real-valued bandpass samples, , are clipped at an amplitude as follows: if if if In the following discussion, we will use a normalized , clipping level, which we call the clipping ratio ( CR where is the rms levelr of the OFDM signal). It is easy to show that, for an OFDM signal with subchannels, for a baseband signal and for a bandpass signal. In and . A CR of 1.4 means our case, the clipping level is about 3 dB higher than the rms level. Filtering after clipping is required to reduce the out-ofband clipping noise. In our simulation, we use an equiripple bandpass finite-impulse response (FIR) filter with 103 coefficients. The stopband attenuation is designed to be 40 dB to guarantee a very low interference level to the neighboring OFDM channels. The passband ripple is 1 dB. III. SIMULATION RESULTS An initial study of the cumulative distribution function (CDF) of the OFDM signal amplitudes [5] reveals that most of the signals are concentrated at low amplitudes, indicating that clipping could significantly reduce the CF with minimal performance degradation. In the following, we concentrate on three performance measures: power spectral density (PSD), crest factor (CF), and bit-error rate (BER). A. Power Spectral Density In this section, our focus is on the spectral splatter caused by clipping and the effect of filtering. The PSD is measured for each OFDM block and then averaged over many blocks to eliminate the effects of the rectangular time window. In Fig. 1, we show the PSD of the clipped signal with various



RTHOGONAL frequency division multiplexing (OFDM) is a very attractive technique for the transmission of high-bit-rate data in a radio environment [1]. However, any multicarrier signal with a large number of subchannels is burdened with a large crest factor (CF peak voltage/rms voltage). When passed through a nonlinear device, such as a transmitter power amplifier, the signal may suffer significant spectral spreading and in-band distortion. The conventional solutions to this problem are to use a linear amplifier or to backoff the operating point of a nonlinear amplifier; both approaches resulting in a significant power efficiency penalty. Several alternative solutions have been proposed to reduce the crest factor (CF) of the signal input to the amplifier [2]–[5]. One of these approaches, and the simplest, is to deliberately clip the OFDM signal before amplification. In particular, since the large peaks occur with very low probability [5], clipping could be an effective technique for CF reduction. However, clipping is a nonlinear process and may cause significant in-band distortion, which degrades the bit-errorrate (BER) performance, and out-of-band noise, which reduces the spectral efficiency. Filtering after clipping can reduce the spectral splatter but may also cause some peak regrowth. Here, we investigate, through extensive computer simulations, the effects of clipping and filtering on the performance of an OFDM wireless communication system.
Manuscript received January 26, 1998. The associate editor coordinating the review of this letter and approving it for publication was Prof. N. C. Beaulieu. This paper was presented at the IEEE Conference on Vehicular Technology, 1997. X. Li is with the Department of Electrical Engineering, University of Washington, Seattle, WA 98195 USA (e-mail: L. J. Cimini, Jr. is with AT&T Laboratories-Research, Red Bank, NJ 077017033 USA (e-mail: Publisher Item Identifier S 1089-7798(98)04029-0.

1089–7798/93$10.00 © 1998 IEEE

However. 2. For CR the out-of-band noise emission power is only 16 dB lower than the signal power. especially at the high percentile points. 2. Power spectral density of the clipped-and-filtered OFDM signals with a clipping ratio of 1. Fig. VOL. From the figure.99%. From Fig.6. Fig. MAY 1998 Fig. B. less than a 1-dB penalty is encountered. Crest Factor Filtering will cause peak regrowth. 3. In Fig. where the CF is also about 9 dB [6]. which causes a degradation in the BER performance.999%.132 IEEE COMMUNICATIONS LETTERS. CR’s from 0. 4. but the question is how much. 2.8 to 1. 99. For ). 1. with forward error-correction coding across the subchannels. This compares favorably clipping (CR with a single-carrier QPSK signal using raised-cosine pulse shaping and a roll-off factor of 0. we show the CF’s of the unclipped OFDM signal (dashed) and the clipped and filtered signal (solid) for five values of the CDF of the signal amplitudes: 90%. we see that filtering greatly attenuates the outof-band emission. This shows that filtering is necessary to suppress the spectral splatter caused by clipping. NO. the degradation is more than 4 dB hard clipping (CR at the 10 BER level. the effect of these ripples should be negligible.999% point is achievable with ) and filtering. the spectral sidelobes after filtering are now at least 50 dB lower than the signal mainlobe. Bit Error Rate (BER) Clipping causes in-band noise.4. Crest factor of the unclipped and clipped-and-filtered OFDM signals. . the out-of-band caused by clipping is evident. In Fig. C. The in-band signal attenuation as well as . with clipping and filtering over a channel of additive white Gaussian noise. 4. Power spectral density of the unclipped and clipped OFDM signals. 3. we show the BER performance as a function of the received signal-to-noise ratio (SNR). 99. and 99. 5. Note that a 9-dB CF at the 99. But when CR . BER of the clipped-and-filtered OFDM signals. Fig. With CR .9%. it is clear that clipping and filtering largely compresses the dynamic range of the signal amplitudes. The 1-dB in-band ripple caused by the FIR filtering may boost the power of some subchannels while suppressing others. averaged across all the subchannels. 99%.5.

[5] X.LI AND CIMINI: CLIPPING & FILTERING ON THE PERFORMANCE OF OFDM 133 IV. pp.” in Proc. no.” in Proc. pp.. pp. 843–847. E. 2.4 and filtering. Li and L. the CF’s at various percentiles of the CDF are used to better characterize the “peakiness” of an OFDM signal. July 1995. [2] T. 21–30. and J. Gross and D. the CF of a bandpass OFDM signal with 128 tones at the 99. [6] T. Wilkinson and A. Mag. With a clipping ratio around 1. pp. 24. Nossek.. May 1997. Cimini. 1634–1638. Lopes. Y. This is at the expense of only a 1-dB degradation in the received SNR. vol. O’Neill and L. Jones. 1995.” in Proc. Daido. A. J. “Envelope variations and spectral splatter in clipped multicarrier signals. 1.999% point is reduced from 13 to about 9 dB. “Effects of clipping and filtering on the performance of OFDM. 71–75.” in Proc. REFERENCES [1] “Special issue on multi-carrier communications. VTC’94. Noguhi. Veeneman. The Netherlands: Kluwer Academic. VTC’95. N. [3] R. A. “Modulation techniques for microwave digital radio. vol. 825–831. “SNR and spectral properties for a clipped DMT ADSL signal. June 1994. [4] R. VTC’97. PMRC’95. 1995. 1986. no. SUMMARY In this letter..” in Wireless Personal Communications. We conclude that clipping and filtering is a promising technique to reduce the CF of OFDM signals using realistic linear amplifiers. Oct. Jr. we investigated the effects of clipping and filtering on the performance of OFDM.” IEEE Commun. which is comparable to the absolute CF of a raised-cosine pulse-shaped bandpass QPSK signal. “Minimization of the peak to mean envelope power ratio in multicarrier transmission schemes by block coding. Sept. Instead of using an absolute CF. . Amsterdam. pp. 10.