Anda di halaman 1dari 6

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 10, NO.

9, SEPTEMBER 2011

2823

Power Optimized PA Clipping for MIMO-OFDM Systems


Hun Seok Kim, Member, IEEE, and Babak Daneshrad, Member, IEEE

AbstractFor a multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) system that is being pushed into power amplier (PA) saturation, this letter investigates power-optimized PA clipping. Our goal is to identify the optimum clipping level for a MIMO-OFDM system that delivers the desired bit error rate (BER) with minimum power consumption in the PA. We present a complete theoretical framework resulting in an analytical expression for the BER of a MIMO-OFDM system subject to PA clipping. PA power saving is addressed by the total degradation metric, which shows that as much as 6dB power reduction can be achieved by proper choice of the clipping level. Index TermsMIMO, OFDM, PA clipping, peak to average power ratio, total degradation.

Transmitter Bit FEC stream Encoder


Spatial Stream Parser

QAM Mapper QAM Mapper

OFDM modulation OFDM modulation

Signal Clipping Signal Clipping

QAM Mapper

OFDM modulation

Signal Clipping Uncoded BER measured here Spatial Stream Deparser

Receiver

OFDM demod. OFDM demod. OFDM demod. MIMO Detector

FEC decoder

Bit stream

I. I NTRODUCTION

Fig. 1.

MIMO-OFDM transceiver structure.

IGH peak-to-average power ratio (PAPR) is a major drawback of OFDM modulation since it results in excessive power consumption at the PA. A number of approaches have been proposed in this area to reduce PAPR for OFDM. Among them, signal clipping which articially limits the amplitude of the signal to a pre-determined clipping level [1] [9] is one of the simplest and extensively studied methods. Although PAPR reduction by signal clipping improves the PA efciency, it also degrades the BER performance by introducing distortion to the desired signal. To quantify this tradeoff space available by various signal clipping strategies, the concept of the total degradation ( ) [3] [6] is widely used. The is a function of the PA amplitude clipping level , dened by ( ) = ( ) + ( ) [in dB] (1) where ( ) is the performance penalty measured by the signal-to-noise ratio (SNR) difference (in dB) between the unclipped system and a system with a nite when both satisfy the same BER. Typically, aggressive clipping (small ) results in lower PAPR at the cost of higher SNR penalty. A conventional system designed to avoid any signal clipping (high ), on the other hand, cannot minimize because of excessive PAPR even if the SNR penalty is 0. There is a rich body of work dealing with the issue of clipped OFDM systems [1] [9]. Analysis provided in these works, however, is focused on single antenna (SISO) OFDM systems and is not readily extended to a MIMO-OFDM system. Although, the work in [9] goes one step further by

analyzing an Alamouti space-time coded system, its result is limited to the single spatial stream case. The total degradation of a MIMO-OFDM system (with multiple streams) has not been fully explored in the literature in part due to the lack of an analytical BER expression for a clipped MIMO system. In this work, we derive analytical BER performance models for a clipped MIMO-OFDM system and identify the total degradation (1) as a function of the clipping level, . As a result, the optimal tradeoff point is revealed for a MIMOOFDM system where the combination of the SNR penalty and PAPR is minimized while a given BER performance requirement is satised. II. S YSTEM M ODEL The structure of the MIMOOFDM system considered in this work is depicted in Fig. 1. Throughout this letter, we use spatial multiplexing transmission with multiple streams. At the transmitter, the information bit stream is parsed to multiple spatial streams. Each stream is then mapped to quadratureamplitude-modulated (QAM) symbols and OFDM modulated. At the receiver, the reverse operations are performed as depicted in Fig. 1. A. Signal Model of a SISO-OFDM System with Clipping We rst present the modeling result from [1] for a clipped SISOOFDM system considering an ideal PA with input and output characteristic given by Input: () = ()() { ()() when () (2) Output: () = () otherwise

Manuscript received October 17, 2010; revised March 21, 2011; accepted June 3, 2011. The associate editor coordinating the review of this paper and approving it for publication was W. Zhang. H. S. Kim is with the Applications and Systems R&D Center, Texas Instruments Inc., Dallas, TX, USA (e-mail: hkim@ti.com). B. Daneshrad is with the EE Department, University of California, Los Angeles, CA, USA (e-mail: babak@ee.ucla.edu). Digital Object Identier 10.1109/TWC.2011.072011.101834

where () and () are input and output signals of the PA respectively and is the input signal power. The amplitude

c 2011 IEEE 1536-1276/11$25.00

2824

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 10, NO. 9, SEPTEMBER 2011

and phase of the input signal at time are denoted by () and () respectively. When the clipping level is set to , the peak PA output signal power is bounded by 2 . It is shown in [1] that, in the frequency domain, the OFDM signal distortion due to clipping is well approximated by a zero-mean Gaussian random variable when the number of OFDM subcarriers, , is reasonably large (e.g., 128). With this assumption, the average PA output transmission power, , after clipping is approximated ([1] [2]) by ( ) 2 = 1 . (3) Using (2) and (3), the PAPR after clipping is presented as a function of by ) ( 2 (4) ( ) = 2 / 1 . The frequency domain PA output signal for the OFDM subcarrier, , can be considered as the sum of the scaled original input signal, , and the Guassian distortion noise signal, , due to clipping [1]. and are uncorrelated and the relationship (5) = + 2 2 holds where = 1 + is the signal power scaling factor (see e.q. (27) of [1]). The distortion signal power for the OFDM subcarrier is denoted by E ( ) = (6)

s as the 1 transmitted symbol vector for the OFDM subcarrier after clipping while the clipping level is common s , can be for all spatial streams. The clipped MIMO signal, represented as the summation of the scaled original transmitted symbol vector, s (without clipping), and Gaussian distortion , as in vector, n s = s + n (8) while E (s n n n ) = 0, E ( ) = I and E (s s ) = I holds. Assuming additive white Gaussian noise (AWGN) at the receiver, we denote the 1 received vector y for the OFDM subcarrier as the product of the channel s plus the matrix H and the transmitted symbol vector 1 AWGN vector n with covariance matrix of E (n n ) = 2 I as in . + n = H s + n y = H s + H n (9) The additive noise plus distortion signal at the receiver is (= H n is + n ). It is worth noting that n denoted by n spatially correlated because of the impact of the MIMO chan . Its covariance, nel on the originally uncorrelated vector n C , is obtained by
2 C = E ( n n ) = H H + I.

(10)

and the distortion signal power power 2 1 . Therefore, is obtained by subtracting the signal power from the PSD as given in } { (1 ) 1 , =0 , 1 , , 1 = [] ( )1 2 2 1 (7)

where the distortion signal power coefcient, , is obtained by calculating the power spectral density (PSD) of the clipped signal and subtracting the signal power from it. For the PSD computation, the upsampled discrete time auto-correlation [] introduced in [1] (see Appendix A of [1]) is adopted. The PSD of the clipped signal (i.e., discrete Fourier transform of) []) consists of the desired signal (
2

We use a frequency-selective quasi-static fading channel model [10], whereby the channel is assumed to be static within the duration of a single packet but is independent from one (,) packet to the next. We denote by the (, ) component of the frequency domain channel matrix H , which is the channel response from the transmit antenna to the receive (,) antenna for the subcarrier. In our channel model, has zero mean independent circularly symmetric, complex Gaussian distribution with unit variance. It is worth noting (,) that has statistically identical distribution regardless of the subcarrier index . It is independent in space and time but correlated in frequency. III. G ENERALIZED MIMO D ETECTION A LGORITHMS FOR THE C LIPPED OFDM S IGNAL In this section, we derive the generalized ML and MMSE MIMO detection algorithms which are applicable to both clipped and unclipped MIMO-OFDM signals. A. ML-MIMO Detection of Clipped MIMO-OFDM Signal The ML solution for the clipped MIMO signal is derived using the conditional probability density function (PDF), (y H , s, ), of the received signal y given channel H , transmit signal s and . The ML decision, s, , is the s that maximizes the conditional PDF as in s, = arg min (y H , s, )
s

where is the over-sampling factor ( 4 is recommended to properly approximate continuous time behavior of the clipped signal [1]) and {} is the output of the -point discrete Fourier transform. It is well known that clipping produces not only in-band distortion but also outof-band interference, which may degrade the performance of other systems in adjacent bands. However, in this work, we investigate the minimum (i.e., lower bound) when spectral growth due to the clipping is allowed and the effect of the outof-band interference is ignored. B. Signal Model of a MIMOOFDM System with Clipping Let and represent the number of receive and transmit antennas. For the MIMOOFDM system, we denote

= arg min
s

2 (y H s) C (y H s) . (2 ) C

(11)

Therefore, s, can be obtained by solving s2(12) H arg min L y L H s2 = arg min y


s s

KIM and DANESHRAD: POWER OPTIMIZED PA CLIPPING FOR MIMO-OFDM SYSTEMS

2825

where L is the Cholesky decomposition of C (= L L ), = L H . The matrix L is essentially y = L y , and H is the whitened signal. the whitening lter and y B. MMSE-MIMO Detection of Clipped MIMO-OFDM Signal The MMSE detection solution for the clipped MIMOOFDM signal, s, , is computed by the product of the received vector y and the MMSE weight matrix W, as in s, = W, y . (13)

simplify the BER estimation by assuming that the minimum distance event dominates the performance. We thus only consider error events which satisfy the relationship, e = s s = [ 0 0 ] , where is the difference between adjacent QAM symbols and e = = is the minimum Euclidean distance between QAM symbols. Based 2 on these assumptions, the system BER given noise power and clipping level is lower bounded by 1 { { }} 1 1 EH Es (s s ) =0 s s.t. s=s { { }} 1 1 EH Es (s s )
=0 s s.t. ss =

The MMSE weight matrix is the W matrix which minimizes the mean square error satisfying { } W, = arg min E (s Wy ) (s Wy ) . (14)
W

By solving (14), W, for the clipped MIMO system is obtained by (( ) 1 ) 2 2 + H H W, = H . (15) + I Applying the non-white noise covariance C (10) to the post detection SNR computing procedure of the unclipped MIMO systems [11] [12], the MMSE post detection SNR of the clipped MIMO system can be obtained in a closed-form expression. We denote by , the MMSE post detection SNR of the clipped MIMO system for the spatial stream of the OFDM subcarrier. The expression for , is given by , = [ {(
E s, s

(17) ) is the

2 where (s s ) = s s H , s, s , , probability of detecting s when s is sent.

A. BER Performance with the ML-MIMO Detector The BER bound for the ML MIMO detector can be computed by applying the ML detection expression for the clipped signal given in (12). To derive the BER bound, we use the fact that, for each symbol in a square -QAM (e.g., 4QAM,) 16QAM, etc.) constellation, there are on average ( 4 4/ adjacent symbols with the minimum distance . Therefore, the BER bound (17) for the ML MIMO detector can be computed by { { 1 1 EH Es =0 s s.t. ss= } }} { 2 2 2 s y s H H H , s, s , , y = 4
1 4 =0

)(

1 s, s

] ) }
,

1 ]
,

= [
2 H I2

((

2 I 2 + H H +

)1
H

1 (16)

where [A], is the diagonal component of the matrix A. IV. A NALYTICAL BER E XPRESSION FOR THE C LIPPED MIMO OFDM S IGNAL An analytical expression for the BER is an important step in quantifying the clipping penalty and (1). We derive an analytical BER expression of a clipped MIMOOFDM system with the ML and MMSEMIMO detection algorithms derived in section III. In this letter, uncoded BER is used as the indicator of the system performance measured at the input of the forward error correction (FEC) decoder as depicted in Fig. 1. The uncoded BER allows the system designer to independently analyze the performance of a clipped MIMO OFDM system regardless of a specic FEC scheme. Typically, there is an one-to-one mapping between uncoded BER and the after-FEC packet error rate, determined by the characteristic of each FEC scheme. Under the assumption of gray -QAM mapping, it is known that the BER and symbol error rate (SER) are related via the approximation; 1 where = 2 ( ) stands for the number of bits per symbol. To estimate the BER, we consider an error event with a correct vector, s, and an error { } vector, s , which satises Es {ss } = Es s s = I. We

{ { { 2 2 Eh E h + L n 2 h
2 h , , ,

L h

(18) } }}

represents a column of H that satises H e = h where h given e. Since only has a non-zero value in either the real or the is a real zero-mean L imaginary part, n h + h L n 2 2 . The h Gaussian random variable with variance 2 2 BER lower bound given and is thus computed as ( ) 1 2 4 4 h (19) , () 2 =0
=0

where () stands for the tail probability function of the normalized Gaussian distribution, and h , () is the probability 2 given and 2 . Note that density function (PDF) of h [ ] 2 = 2 H C1 H h is independent and identically , distributed regardless of the spatial stream index .

2826

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 10, NO. 9, SEPTEMBER 2011

T HE PDF OF 1/

TABLE I [(

H H + 0 I

)1 ]
,

0 , 22 0 32 0 42 0 33 0 43 0 44 0

PDF of 1/

[(

( ( ( ( ( (

) ) ) ) ) ) (

H H + 0 I

)1 ]
, 1 (+1)2

0 0 1 + 0
0 1 + 2 0 2 0 3 0

1 given 0 > 0
0 +1

0 2 0 6

1 2(+1)2 1 6(+1)2

0 +1 2(+1)

) )
2 0 +30 +1

0 0 1 + 20 +
0 1 + + 2 0 0 32 0 2

1 2

+
2 0 2

0 +2 6(+1)

0 +2 (+1)3

2 0 6 2(+1)2

) )
2 3 0 +80 +120 2(+1)

2 0 6

0 +3 3(+1)3

2 0 +20 6 6(+1)2

2 0 +50 +3 3(+1)

0 0 1 + 30 +

3 0 6

2 0 +60 +6 2(+1)4

3 0 +360 48 6(+1)3

2 3 0 +60 12 2(+1)2

2 using the result We derive the exact PDF, [ h , (), of h ] in [11] where the PDF of 1/ (H H + 0 I)1 , 1 is studied for a positive constant 0 and a matrix H with independent complex Gaussian components. When H is an complex Gaussian matrix as in our channel ] [ with unit variance model, the PDF of 1/ (H H + 0 I)1 , 1 is represented by the function in TABLE I which was originally derived for the unclipped MMSE MIMO system [11]. In order to nd h , (), we use the matrix inversion lemma, 1 1 (H H + I ) = 1 H H , which leads 0 (H H + 0 I) 0 us to relationship
2 2 2 1 2 ( )2 h [( )1 ] 2 = 1/ H H + I

TABLE II E XAMPLE M ODES (PARAMETER S ETS )

Mode Mode1 Mode2 Mode3 Mode4 Mode5 Mode6

23 24 34 44 22 33

Constellation 64QAM 16QAM QPSK 64QAM 16QAM QPSK

MIMO Detection ML ML ML MMSE MMSE MMSE

(20) 1.

OFDM subcarrier is an independent AWGN channel whose SNR is dictated by the post-detection SNR, , , given in (16). In this case, the BER lower bound (17) for the MMSE detector can be computed by

Since the PDF of the random variable on the right side of (20) 2 is obtained by random is , the exact PDF of h variable transformation and is given by ( )2 ( 1 2 2 ) 2 h (21) , () = 0 = when > 0 2 ( 2 ) 1 otherwise. ( 1)! It is obvious that when the clipping distortion is negligible ( [ 1 or , () is identical to the ] 0), h 2 H H , which has the Gamma (Chi-square) PDF of distribution (21). The plot of h , () is shown in Fig. 2, which indicates that our expression (21) is well matched to the Monte-Carlo simulation results for various , and clipping levels. B. BER performance with the MMSE-MIMO detector The output of an MMSE-MIMO detector is the equalized , satisfying symbol vector, s , along with the additive term, n the equation, W, y = s + n , where n represents the post-detection noise-plus-interference. In general, for the MMSE-MIMO detector, symbol-by-symbol detection [11] [12] is performed assuming the spatial channel of the

1 4 =0

{ { EH Es

} )}

s s.t. ss=

s W, y 2 s W, y 2 = = 4
1 4

(22) { { [ ]}} 2 2 EH Ee n e n { [ ]} 2 EH 2{ } H , ,

=0 1 4 =0

where is a component of n that satises n e = 2 given e. Note that given H , and is a zero-mean complex Gaussian random variable whose variance is dictated by 1/, . The MMSE post-detection SNR, , , (16) has independent, identical distribution regardless of the spatial index . We de2 and . Applying the note by , () the PDF of , given matrix inversion lemma and random variable transformations

KIM and DANESHRAD: POWER OPTIMIZED PA CLIPPING FOR MIMO-OFDM SYSTEMS


-4

2827

x 10

10

or p ,k ( x)

10

Monte-Carlo Simulation Analysis

Target BER = 10-4

NT ML,

3, N R 1.2

4,

2 N

0.0316,
Total Degradation (dB)
8

Target BER = 10-3

NT

2, N R
NT

2, 1.3

2 N

0.01,

ph ,k x

MMSE ,
6

2, N R 1.7
2 N

3,

2 N

0.0032,

PDF value

ML,
4

NT
2

4, N R

4, 1.7

0.0032,
4
Target BER = 10-4

MMSE ,

Mode1 Mode2 Mode3 Mode4 Mode5 Mode6 2.5 3

0.5

1.5

2
Clipping level

20

40

60

80

100
i ,k

120

140

160

Value of

hk

Clipping Level

(for ML) or

(for MMSE)
Fig. 4. Total degradation to satisfy 103 or 104 BER.

Fig. 2. Validation of h , () and , (). Comparison between MonteCarlo simulation and analytical PDF.

C. BER Analysis Validation


Simulation Analysis

Mode2
-2

Mode4

10 BER

Mode3 Mode1

Mode5

1.2

1.4 1.7

1.3

1.7

10

-3

10

15

20

25 30 SNR dB

35

40

45

50

Fig. 3.

BER analysis vs. simulation results for various clipping.

to (16), we arrive at exact the expression ) ( 2 2 2 / 2 + , () = ( ( ))2 2 2 1 + ( + 1) 2 2 + (


2 2 (+1)2 /(2 + )

) . (23)

1+(+1)

2 2 2 2 +

) 0 =

2 2 + 2

To validate the BER analysis, we compared Monte-Carlo simulations with the results predicted by (19) and (24). Fig. 3 shows a side by side comparison of the BER performance of a collection of representative parameter sets itemized in TABLE II. For BER simulation, we used a MIMOOFDM system with =128 subcarriers and 20 OFDM symbols per packet without any pilot or training tones. The channel model described in section II.B was applied with perfect channel estimation and synchronization. Fig. 3 conrms that the BER lower bound given by (19) and (24) is valid and reasonably close to the actual BER for various clipping levels. For the ML detector (Mode 1 3), the gap between the actual performance and our analytical expressions is mainly because of looseness in the lower bound (17), where we only consider minimum distance error events ignoring spatial interference. The bound (17) is much tighter for MMSE cases (Mode 4 6) because the MMSE detector is designed to minimize spatial interference before making decisions. It is worth noting that the total degradation (1) is governed by ( ) (SNR gap between the unclipped and the clipped cases) rather than the exact position of the BER curve. Since the BER performance penalty due to clipping is accurately predicted by our analytical model, we will use (19) and (24) with equality to compute the SNR penalty of clipping in total degradation (1). V. PA P OWER S AVING VIA T OTAL D EGRADATION A NALYSIS The tradeoff space of clipping is quantied by the total degradation ( ) metric [3] [6] as a function of dened in (1), which combines the performance degradation penalty and the PAPR penalty. The objective of the optimal clipping is to minimize the metric directly related to the PA power consumption to achieve a given target BER performance. The SNR degradation, ( ), in (1) can be accurately found by evaluating the inverse function of the BER lower bound expressions (19) or (24). Since the inverse function of

Fig. 2 shows that our expression (23) for , () is well matched to the Monte-Carlo simulation results for various clipping levels. Finally, using the expression for , (), we rewrite the lower BER bound (22) for the MMSE detector and obtain the expression ) ( 1 2 4 4 . (24) , () 2 =0
=0

2828

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 10, NO. 9, SEPTEMBER 2011

(19) or (24) is difcult to obtain, we compute ( ) by numerically evaluating (19) or (24) for a target BER. Fig. 4 shows the penalty as a function of the clipping level to satisfy a target uncoded BER (at the input of FEC decoder) of 103 or 104 for the parameter sets itemized in TABLE II. The optimal clipping level for each parameter set is the which minimizes the on each curve. Fig. 4 shows that the values resulting from the optimal clipping were in the range of 3.2dB (for Mode6) to 7.6dB (for Mode1) depending on the parameter set. Meanwhile, the of the conventional unclipped system with 3.1 is at least 9.4dB (this is solely from the PAPR penalty since, by denition, the SNR penalty is 0 for unclipped systems). In other words, the optimal clipping can provide the same BER performance with 1.8 6.2dB less penalty (SNR and PAPR penalty combined) compared to the unclipped system. Fig. 4 suggests that the optimal clipping level is signicantly dependent on the operating system parameters such as the MIMO detection algorithm, QAM size, MIMO antenna conguration, etc. It is also worth noting that modes with smaller constellations (e.g. Mode6 for MMSE or Mode3 for ML) can extract more gain from optimal clipping since these modes operate at lower SNRs and thus more aggressive clipping can be tolerated until the clipping distortion starts to dominate the system performance. VI. C ONCLUSION We identied optimal PA clipping for MIMO-OFDM systems as a means of minimizing PA power consumption for a target BER. The BER performance of a clipped MIMOOFDM signal was successfully analyzed and an accurate BER approximation was provided for various combinations of system parameters. Our studies show that the total degradation

penalty of the optimized system can be 1.8 6.2dB lower than an unclipped system satisfying the same BER requirement. R EFERENCES
[1] H. Ochiai and H. Imai, Performance analysis of deliberately clipped OFDM signals, IEEE Trans. Commun., vol. 50, no. 1, pp. 89101, Jan. 2002. [2] , On the distribution of the peak-to-average power ratio in OFDM signals, IEEE Trans. Commun., vol. 49, no. 2. pp. 282289, Feb. 2001. [3] D. Dardari and V. Tralli, Analytical evaluation of total degradation in OFDM systems with TWTA or SSPA, IEICE Trans. Commun., vol. 85, no. 4, pp. 845848, 2002. [4] P. Banelli, Theoretical analysis and performance of OFDM signals in nonlinear fading channels, IEEE Trans. wireless Commun., vol. 2, no. 2, 2003. [5] P. Banelli, G. Baruffa, and S. Cacopardi, Effects of HPA nonlinearity on frequency multiplexed OFDM signals, IEEE Trans. Broadcast., vol. 47, no. 2, pp. 123136, 2001. [6] S. C. Thompson, J. G. Proakis, and J. R. Zeidler, The effectiveness of signal clipping for PAPR and total degradation reduction in OFDM systems, IEEE Globecom Conference, 2005. [7] E. Costa, M. Midrio, and S. Pupolin, Impact of ampliers nonlinearities on OFDM transmission system performance, IEEE Commun. Lett., vol. 3, pp. 3739, Feb. 1999. [8] R. J. Baxley and G. T. Zhou, Power savings analysis of peak-to-average power ratio reduction in OFDM, IEEE Trans. Consum. Electron., vol. 50, no. 3, pp. 792798, Aug. 2004. [9] H. A. Suraweera and J. Armstrong, Error performance analysis of clipped Alamouti space-time coded OFDM systems, in Proc. TENCON 2005, pp. 16, Nov. 2005. [10] I. Lee, A. M. Chan, and C.-E. W. Sundberg, Space-time bit-interleaved coded modulation for OFDM systems, IEEE Trans. Signal Process., vol. 52, no. 3, pp. 820825, Mar. 2004. [11] N. Kim, Y. Lee, and H. Park, Performance analysis of MIMO system with linear MMSE receiver, IEEE Trans. Wireless Commun., vol. 7, no. 11, pp. 447478, Nov. 2008. [12] M. Butler and I. Collings, A zero-forcing approximate log-likelihood receiver for MIMO bit-interleaved coded modulation, IEEE Commun. Lett., vol. 8, no. 2, pp. 105107, Feb. 2004.

Anda mungkin juga menyukai