Sie sind auf Seite 1von 5

PAPR and Other Measures for OFDM Systems with Nonlinearity

Ali Behravan and Thomas Eriksson

Department of Signals and Systems


Chalmers University of Technology, SE 412 96 Gothenburg, Sweden
ali.behravan@s2.chalmers.se, http://www.s2.chalmers.se

Abstract  S P
  
Input bits
Baseband / IFFT Nonlinear
In this paper, we discuss the effects of nonlinearity on the Modulator / Amplifier
N points
P
performance in transmission of OFDM signals. Further- S

more, we study measures related to nonlinearities, such


Channel
as PAPR (Peak-to-Average-Power-Ratio), with respect to
their ability to correctly predict the effect of a nonlinear-   P
S
Output bits
Baseband
ity in an OFDM system. We also propose new measures PSfrag replacements
Demodulator
/ FFT /
S N point
for signal distortion in nonlinear systems. The perfor- P

mance of the measures are studied using several exam-


ples.
Figure 1: Block diagram of an OFDM system
Keywords
Orthogonal Frequency Division Multiplexing (OFDM),
Nonlinearity, Peak-to-Average Power Ratio (PAPR). In other words: does a reduction of PAPR always lead to
a decrease in the effects of the nonlinearity?
1. INTRODUCTION In some recent contributions [8], [9], other measures
Due to its robustness against time dispersion in multipath have also been mentioned for the sensitivity of multicar-
fading channels, OFDM is one of the major candidates rier systems to nonlinearity. In this paper we study the
for the future high speed wireless data communication PAPR and also measures related to the power of signal,
systems. In an OFDM system the data is transmitted over and compare their performance using several examples.
a number of parallel frequency channels, each being mod- The remainder of this paper is organized as follows.
ulated by a baseband QAM or PSK symbol. As a result In Section 2 we first study the nonlinearity in an OFDM
the amplitude of such a signal can have very large values. transmission system and define the distortion as the real
This large amplitude variations is a big disadvantage of effect of nonlinearity in the system. Then in Section 3 we
OFDM, which tends to reduce the power efficiency of the discuss different measures corresponding to signal vari-
RF power amplifier. Since the amplifier is typically rather ations. In Section 4 the capabilities of different mea-
nonlinear for high input amplitudes, it may be necessary sures in predicting the effect of nonlinearity on the system
to decrease the input amplitude to reduce the distortion. performance are quantified and compared. Finally some
The effect of nonlinearities on OFDM transmissions conclusions are drawn in section 5
has been studied in different contributions [1], [2], [3].
It has been shown that when the nonlinearity is known, 2. OFDM SIGNALS IN NONLINEARITY
one can get the best amplifier efficiency, by careful se-
lection of the input signal power. However, a system de- Fig. 1 shows a basic block diagram of an OFDM sys-

signer must choose the baseband signal set to minimize tem. The baseband modulated symbols are first passed
the distortion. Therefore a measure should be devised to through a serial to parallel converter, forming a complex

quantify the signal variations and nonlinear distortion. vector of size . Then each vector is transformed by
A well known measure of the amplitude variations means of an IFFT to an OFDM symbol which will be
of an OFDM signal is Peak-to-Average Power Ratio the channel symbol. The nonlinear block at the output of
(PAPR). A number of approaches have been proposed the modulator represents the high power RF amplifier to-
to reduce PAPR of an OFDM signal [4], [5], [6], [7] gether with the digital to analog converter. At the receiver
thereby seeking to minimize the effects of the nonlinear- the inverse of the above operation is done to form the set
ities. However, it is not entirely clear how well the PAPR of decision variables.
measure is related to the real effects of the nonlinearity. The OFDM modulated signal 
is usually modeled
1
as the sum of independent random variables. According 10

to the central limit theorem if the number of subcarriers


is large, the signal can be approximated as a Gaussian
distributed random variable. Using Bussgangs theorem
the signal at the output of nonlinearity can be written as

Distortion to Signal Ratio


2
10
the sum of an attenuated input replica and an uncorrelated
distortion term [1], [10]
    (1) 

where the complex gain  describes attenuation and


3
10

 

   
phase rotation, and can be derived as


 
(2)
 is the distortion term, which is uncorrelated with
 PSfrag replacements 4
10
0 1 2 3 4 5 6 7 8

IBO

and approximated as a Gaussian noise term. The auto-


Figure 2: Distortion to Signal Ratio vs. IBO
           
correlation function of the output signal can be written
as
(3)
Equation (3) can be used to derive the power of distortion where max represents the clip level. V
WPXZY
for different subcarriers. The Distortion to Signal power Ratio at the receiver
At the receiver, the output of the FFT block gives a is derived using equation (7). Fig. 2 shows as a
function of Input Back-Off (IBO) for a 1024 OFDM sys-
      
set of decision variables
tem when a 16QAM baseband modulation is used. By
(4) IBO we mean the amount of decrease in the input signal

   !# " ('*),+.-0/(132 4.5,6 7


power compared to the power at the clip level. Since DSR
where
has a direct translation in the performance of system we

%$&
   (5) use it as a measure for the sensitivity of the OFDM signal
to a known nonlinearity.
and
!8"
    #  '9),+-0/(1 2%4.56 7
3. SIGNAL MEASURES

%$8&  (6) As mentioned earlier, amplitude limiting introduced by a


nonlinearity degrades the performance of an OFDM sys-
Therefore the equivalent linear model of the OFDM tem. A measure of the degradation can be very helpful in


transmission with nonlinearity consists of a complex gain evaluating the performance of a given system, and in de-
, and an uncorrelated additive Gaussian distortion [11], signing a signaling set that avoids degradation. One such
[12]. The performance of this system can then be eval- measure is PAPR, defined as
uated in the same way as an AWGN channel . The bit
error rate of an AWGN system is well-known to be an in-
[V[]
Y \_^akB`%l bdcfehgic j (9)
creasing function of the Distortion to Signal power Ratio c%e g c jmon
!8"
       :<  ;>=   @  %$!&B8"A CE D
(DSR). In this case DSR is At a symbol level this means that we have to look at the
maximum value among the samples of a symbol ( ).
p eg
3: ?%=  @ %$8& :3F D (7)

Theoretically this maximum can get as high as , but the

 probability of occurrence of such a peak is very small,


where A C D and : F D are the variance of distortion and


especially for high number of subcarriers. In practice in

6
order to decrease the intercarrier interference and out-of-
power of signal in the th subchannel, respectively. If the
band radiation it is required that the amplifier operates in
input-output characteristic of the nonlinearity is known,
its linear region. Therefore it is desirable to limit the max-
one can find the corresponding Distortion to Signal Ratio
imum envelope of the multicarrier signal, namely PAPR
using (7). Here we take the example of a soft-limiter type
of the signal. On the other hand reducing the PAPR does
of nonlinearity, where the input-output relation is
not necessarily mean an improvement in the performance

 K LI N/ M
HG - 7 J max
 POQ

SR /NM M max of system.
However, there are other possibilities in defining a
M PTUM
    max (8)
max  max measure. A measure of the signal degradation in non-
linearity should have a set of desirable properties:
16

0.34
It should be independent of the nonlinearity.
14
0.32
It should be easy to compute.
12 0.3
It should be highly related to the real effects of a

Excess power
nonlinearity.

PAPR
0.28
10

The first property is desirable since the nonlinearity and 0.26

the operating point of the system is typically unknown 8

at the time of baseband system design. The second one 0.24

is important in the sense that the measure must be easily 6


0.22
attainable using a limited set of data. The third one is
important for obvious reasons. PAPR measure fulfills the PSfrag replacements
4 0.2
first requirement. When considering the PAPR of a dis- 0.05 0.055 0.06 0.065
Distortion power
0.07 0.075 0.05 0.055 0.06 0.065
Distortion power
0.07 0.075

crete signal it fulfills the second property as well. How-


ever, according to (7), the performance of an OFDM sys- Figure 3: PAPR and Excess power as functions of distor-
tem with nonlinearity depends on the power of distortion, tion
while the peak value of a signal does not show the power
of the signal or the distortion. This may make the the
PAPR measure less suitable in some applications.
We define excess power of the OFDM symbol as the
4. SIMULATIONS
amount of power above a certain power level, In this section we present some simulation results show-

: excess
   -   /   7  -   T   7 
    (10)
ing the relations between the distortion and the measures
that were introduced.
 
where  is the average envelope of the signal.
We set up a simple experiment with an OFDM system
using 1024 subcarriers, a 16QAM baseband modulation

:C
The above definition is one example of a large set of and a soft limiter type of nonlinearity. As discussed in
measures that can be used for amplitude variations. A the previous section, the distortion power shows the
more general form of (10) can be written as effect of a known nonlinearity on the performance of the

 :C
system. The quality of the two discussed signal measures

:
      -   /  7 -   T 7
:C
can therefore be evaluated using . Fig. 3 shows scatter
:
:C
excess   (11) plots for excess versus , and PAPR of the signal ver-

   
where is the reference level for the distortion (in equa-
:
sus . Both PAPR and excess are computed over single
OFDM symbols. It can be seen from the figure that for

:C
tion (10), 
.) the example used here, PAPR of signal is less related to
As will be shown in the next section the above set of
definitions are all independent of the type of nonlinearity.
than excess . :
For high back-off values, it is very likely that the dis-
In addition they are easy to compute and we will show tortion is because of only one sample that crosses the clip
:  

that the excess is highly related to the distortion for some
values of .
level. Therefore the PAPR is a good measure of nonlinear
distortion in high back-off values. This effect is shown

 
 
It is straight-forward to show that excess gives the
peak of signal when . Hence
: 
 in Fig. 4, where the correlation coefficients between the
distortion power and PAPR, and distortion power and ex-
cess power are plotted for different values of back-off. It
!
" "*)
)  
can be seen that when the nonlinearity is severe, excess
:M: $#&%
: ! ' 
excess ( O (12) power can show the distortion very well, while for high
back-offs PAPR is a better measure.

: 

In the next example we compare two algorithms to re-

  2
On the other hand excess can be a measure of power duce the distortion. One is based on PAPR reduction and
when 
, as defined in (10). the other is based on the reduction of excess . We take :
:
Both PAPR and excess fulfills the two first proper-
 

an OFDM system with number of subcarriers each be-


ties above. The question is therefore how well they fulfill ing modulated by a BPSK symbol. Assume that a rate
5+ 6 - / 2 : 7
the third property. In the next section we answer this by code is used to reduce the PAPR and excess ac-

means of simulation of an OFDM system with nonlinear- cordingly. This code is able to discard portion
ity. of the total number of OFDM symbols, with the highest

1
PAPR reduction scheme
excess reduction scheme
0.9 N=1024

0.06
0.8

0.7
Correlation coefficient

Distortion to signal ratio


N=64
0.6
0.055
PSfrag replacements
0.5

Rejection Rate
0.4 0.05
Distortion
0.3

0.2
0.045

0.1

N=8
PSfrag replacements
0
0 1 2 3 4 5 6 7 8 0.04
IBO (dB) 0 1/2 3/4 7/8 15/16 31/32

:C
:C
Figure 4: Correlation coefficients between PAPR and Figure 6: Distortion to signal power ratio as a function of
(solid line) and  
back-off
and (dashed line) vs. input: F0F rejection rate

:
It is also interesting to look at the distortion as a func-
PAPR and excess , correspondingly. tion of ratio of rejected symbols. In Fig. 6, the distortion
Since we do not propose any coding scheme here, the for the two cases plotted for different rates of rejection.
following illustration is used to show the effect of coding In accordance with Fig. 5, for different symbol rejec-
on reducing the distortion. Let us assume that two coding
schemes have been devised that are able to reject the sym-
:
tion rates, the excess rejection method gives lower dis-

:
tortion than the PAPR reduction scheme. It can also be
bols with the highest PAPR and excess , respectively. Fig. seen from the figure that the distortion in the system with
5 shows the resulting distortion based on the two meth- lower number of subcarriers decays faster, with the rejec-
 
ods when half of the symbols are rejected. The number
  tion rate. This is because of the fact that a certain rejec-
2

of subcarriers in this example is . As the fig- tion rate is equivalent to different coding rate for systems
ure illustrates, the PAPR reduction method has a higher
:
with different number of subcarriers.
distortion compared to the reduction based on excess . We
"" & 
note that this ratio of rejection is equivalent to using a
code with rate
. This high rate explains the small
difference between the curves.
& 5. CONCLUSIONS
In this paper we studied the effect of nonlinearities on the
performance of OFDM systems. The relevance of PAPR
0.06
Without symbol rejection
PAPR reduction scheme
excess reduction scheme
:
and excess with the nonlinear distortion were also dis-
cussed. The suitability of the new measure in describing
the nonlinear distortion was illustrated by simulations.
0.05
The new measure is easy to compute and close to the real
distortion caused by nonlinearity. Several examples were
Distortion to signal ratio

0.04
also presented to compare the performance of system us-
0.03
:
ing codes based on excess and PAPR reduction methods.

Acknowledgment
0.02

The first author wishes to acknowledge the fund received


0.01 from the PCC++ program.
PSfrag replacements
DSR
0
0 0.5 1 1.5 2 2.5 3 3.5 4 6. References
IBO (dB)
[1] P. Banelli and S. Cacopardi Theoretical analy-
Figure 5: Distortion to Signal power Ratio vs. Input sis and performance of OFDM signals in nonlinear
Back-off AWGN channels, in IEEE Transactions on Com-
munications, vol. 48, no. 3, March 2000, pp. 430
441.
[2] H. Ochiai and H. Imai Performance analysis of de- [8] R. Neil Braithwaite Using Walsh code selection to
liberately clipped OFDM signals, in IEEE Trans- reduce the power variance of band-limited forward-
actions on Communications, vol. 50, no. 1, Jan. link CDMA waveforms, in IEEE Journal on Se-
2002, pp. 89101. lected Areas in Communications, vol. 18, 2000, pp.
22602269.
[3] D. Dardari, V. Tralli and A. Vaccari. A theoreti-
cal characterization of nonlinear distortion effect on [9] A.R.S. Bahai, M. Singh, A.J. Goldsmith and
OFDM systems, in IEEE Transactions on Com- B.R. Saltzberg A new approach for evaluating clip-
munications, vol. 48, no. 10, Oct. 2000, pp. 1755 ping distortion in Multicarrier systems, in IEEE
1764. Journal on Selected Areas in Communications, vol.
20, 2002, pp. 311.
[4] J.A. Davis, J. Jedwab Peak to mean power control
in OFDM, Golay complementary sequences, and [10] A. Papoulis Probability, random variables, and
Reed-Muller codes, in IEEE Transactions on In- stochastic processes, 3rd ed. New York: McGraw-
formation Theory, vol. 45, Nov. 1999, pp. 2397 Hill, 1991.
2417.
[11] A. Behravan, F. Munier, T. Svensson, M. Flament,
[5] A.E. Jones, T.A. Wilkinson and S.K. Barton Block T. Eriksson, A. Svensson and H. Zirath System im-
coding scheme for reduction of the peak to mean plications in designing a 60 GHz WLAN RF front-
envelope power ratio of multicarrier transmission end, in Proc. GHz2001 Symposium, Lund, Swe-
schemes, in Electron. Lett., vol. 30, 1994, pp. den, Nov. 2001.
20982099.
[12] E. Costa and S. Pupolin M-QAM-OFDM system
[6] K.G. Paterson Generalized Reed-Muller codes and performance in the presence of a nonlinear amplifier
power control in OFDM modulation, in IEEE and phase noise, in IEEE Transactions on Commu-
Transactions on Information Theory, vol. 46, 2000, nications, vol. 50, 2002, pp.462472.
pp. 104120.
[7] V. Tarokh and H. Jafarkhani On the computation
and reduction of the peak-to-average power ratio in
Multicarrier communications, in IEEE Transac-
tions on Communications, vol. 48, 2000, pp. 3744.

Das könnte Ihnen auch gefallen