Sei sulla pagina 1di 5

Peak-to-Mean Envelope Power Ratio Statistical Analysis of Continuous OFDM Signal

Yves Lou t, Sajjad Hussain e SUPELEC/IETR - Campus de Rennes Avenue de la Boulaie, CS 47601, F-35576 Cesson-S vign Cedex, France. e e Phone: + (33) 2 99 84 45 00, Fax: +(33) 2 99 84 45 99 Yves.Louet, Sajjad.Hussain@supelec.fr

Abstract Peak-to-Mean Envelop Power Ratio (PMEPR) statistical analysis in continuous OFDM context is addressed. After having precisely dened this term which characterizes OFDM signal uctuations, rst statistical order derivation of PMEPR from two already developed distribution functions for continuous base-band OFDM signals is proposed. In literature, one of these two functions is simulation based and lacks any theoretical justications. It is demonstrated that rst statistical order of these two distributions is almost same which provides the analytical justication of the correctness of the simulation based distribution function. Moreover, it has been revealed that the second statistical order of the PMEPR valid for an oversampling factor of one (discrete signal), could also be adopted for continuous signals. PMEPR distribution function associated with continuous OFDM signal is proposed. Index Terms PMEPR, Probability Density, CCDF, OFDM.

I. I NTRODUCTION rthogonal Frequency Division Multiplexing (OFDM) modulation suffers from large uctuations of the temporal RF signal. This leads to major difculties for power amplication causing spectral regrowth and in band distortions if no sufcient Input Back Off (IBO) is taken. This high PMEPR problem has generated huge quantity of methods aiming to reduce PMEPR and/or to increase the High Power Amplier (HPA) efciency. The other side of the problem concerns the best theoretical description of PMEPR. As a rst approximation, some authors have proposed upper bound PMEPR denition by considering the maximum of the instantaneous power[5]. The worst case has been considered for which all the symbols are added in phase, producing the highest signal amplitude. Consequently upper bound is attained, which is proportional to the multiplex carriers number. This upper bound is rare to achieve, especially when the number of carriers is large. This deterministic approach can not match reality and a statistical behavior description of the PMEPR is preferable as PMEPR itself happens to be a random variable. Considerable research has been done on this subject, leading to some Cumulative Distribution Functions (CDF) denitions of the PMEPR. The well-known OFDM Gaussian-like behavior hypothesis is veried when the carrier number is greater than 64. It has lead to a CDF expression, valid for an oversampling factor equal to one [1]. This Gaussian assumption is related to the complex base-band discrete OFDM signal. But as

the amplied signal is continuous, this CDF is not valid for a high oversampling factor (continuous signals). Then, CDF expressions have been proposed to approximate the continuous signal PMEPR distribution. The original approach was the application of the level crossing rate theory to OFDM signals. The bounds proposed by [2] and [3] are very close to simulations for oversampling factors equal to four and greater, simulating nearly a continuous signal. Also [1] presents the distribution function for continuous signals but it is important to mention that this bound is simulation based without theoretical justications. Even if cumulative distribution functions describe greatly the statistical developments, it is of major importance to evaluate the associated statistical orders. In [5], the authors propose further developments from the distribution function of [1] and get the expressions of the rst and second statistical orders versus the number of subcarriers in an OFDM symbol. These expressions are derived from the distribution function obtained with an oversampling factor of one. Thus do not seem to be valid for continuous OFDM signals. In this paper, we analyze this point by giving valid expression for the rst statistical order of the PMEPR for continuous OFDM signal with the help of [2]. The paper is organized as follows: Section II describes Power Ratio, Cumulative Distribution Functions of PMEPR and PAPR. Section III presents the new rst statistical order derivation for continuous signals. Results are given in section IV. Finally, conclusions are drawn in Section V. II. C UMULATIVE D ISTRIBUTION F UNCTIONS OF PMEPR AND PAPR A. Elementary PMEPR and PAPR Denitions 1) Peak to Mean Envelope Power Ratio (PMEPR): Let Xk be the k th complex modulation symbol from any digital modulation and let X be the vector of N symbols. The N discrete OFDM signal samples xn , n = 0, ..., N 1 are the result of the Inverse Fast Fourier Transform (IFFT) of the vector X. This output vector x is said to be an OFDM symbol of N subcarriers whose components are given by
N 1

x(n) =
k=0

Xk e2j

kn N

, n = 0, ..., N 1.

(1)

978-1-4244-1645-5/08/$25.00 2008 IEEE

1681

Fd () = P r(P M EP Rd (x) ) (1 e )N .

(7)

Fig. 1.

Illustration of PMEPR and PAPR localization.

In practice, the IFFT is followed by an oversampling of factor L, a Digital to Analog Converter (DAC) and a low pass lter. By assuming the independency of the Xk symbols, each symbol following a uniform law, one may consider that the real (I) and imaginary (Q) parts of x are asymptotically Gaussian processes for large N values. Then, one may think that the probability of I and Q to reach high values is not negligible which implies an analysis on the temporal uctuations of x. For an entity z, the classical metric for signal uctuation description is the Power Ratio [6] dened by PR = M ax{|z|2 } , E{|z|2 } (2)

The last inequality stands for the PMEPR distribution function denition when the maximum values are considered; not the effective values. This upper bound is a good approximation of the P M EP Rd of x with oversampling rate of one but differs from the P M EP Rc distribution of x(t). We know that when the oversampling factor L tends to innity, the signal n n x( L ) tends to x(t). Then, the P M EP Rd of x( L ) approaches to P M EP Rc of x(t) for large L values. Being rigorous, we have to consider the low pass lter inuence on the signal uctuation distribution but without loss of generality, we assume in this paper that the lter induces no peak regrowth. 2) Base-Band Continuous Signals Case: In [2] and [3], a CDF upper bound has been proposed for high oversampling factor values, simulating nearly the continuous signals, what leads to the CDF Fc denition of P M EP Rc , Fc () = P r(P M EP Rc (x) ) e
3

where E{.} is the expectation operation. Here, z can be any signal; discrete, continuous, base-band or RF. This P R denition is supposed to be general enough to take most of the well known denitions of power uctuations into account. For example, the PMEPR is the PR derivation for baseband signals, discrete or continuous. For simplicity, we will adopt the following notations : P M EP Rd and P M EP Rc for discrete and continuous base-band signals respectively, with the mathematical denitions, P M EP Rd (x) = M axn[0,N ] |x(n)| , E{|x(n)|2 }
2

(8)

III. PMEPR S TATISTICAL O RDERS From the last session, it can be derived that the probability density function related to Fd function is fd () = N e (1 e )N 1 . (9)

(3)

Then, after some mathematical developments, it can be shown [5] that the rst associated statistical order md for large N values can be written as md =
+

for the PMEPR of a discrete signal x(n) of N samples and M axt[0,T ] |x(t)|2 , P M EP Rc (x) = E{|x(t))|2 } (4)

fd ()d lnN + ,

(10)

for the PMEPR of the continuous signal x(t) of duration T . 2) Peak to Average Power Ratio (PAPR): The PAPR is a metric dedicated to describe RF signal uctuations. PAPR of RF signal s(t) with carrier frequency f0 , P AP R{s(t)} = )| M ax|Re(x(t)e . j2f0 t )|2 } E{|Re(x(t)e
j2f0 t 2

where is the Euler constant ( 0.577). 2 In the same way, the second statistical order d for large N values [5] is given by
2 d = +

2 fd ()d

2 . 6

(11)

(5)

The relationship between PMEPR and PAPR is demonstrated in Fig. 1. B. Cumulative Distribution Functions 1) Base-Band Discrete Signals Case: Assuming ergodic processes and independence between the N samples of x, it can be shown that the cumulative distribution function F of |x| is given by F () = P r(|x| ) = (1 e 2 )N ,
2

(6)

where 2 = E{|x|2 }. A simple change of variables yields the CDF Fd of the P M EP Rd ,

The approximations provided by (10) and (11) are very close to simulation results as shown in Tables I and II. These simulations are performed for QPSK modulated 105 OFDM symbols with an oversampling factor L = 1 and different values of OFDM subcarriers N . Tables III and IV recount results for OFDM signal with an oversampling factor L = 4 (here L = 4 almost simulates a continuous PMEPR estimation [4]). Two conclusions can be drawn from Tables I to IV : rstly, Eq.(10) is no longer valid for an oversampling factor greater than one; secondly, Eq. (11) still ts quite well with simulation results and can denitely be adopted for the PMEPR variance estimation in a continuous signal context. Next section will discuss the derivation of a new mean PMEPR expression for continuous base-band OFDM signals.

1682

A. Mean PMEPR Expression


TABLE I M EAN VALUES COMPARISON BETWEEN SIMULATIONS (ms ) AND (10) FOR L=1 N 64 128 256 512 1024 2048 4096 ms 4.8 5.47 6.15 6.83 7.52 8.2 8.9 Eq.(10) 4.73 5.42 6.12 6.81 7.5 8.2 8.89

1) Theoretical Developments : By denition, rst statistical order is related to density function fc () = d(Fc ())/d by mc (N ) =
+

fc ()d.

(12)

Starting from Eq.(8), the probability density function fc associated to Fc veries fc () = N 2 (1 2) 3 e e[


3

e ]

(13)

The rst statistical order mc vs N is then equal to the sum of two terms A and B, where
TABLE II
2 VARIANCE VALUES COMPARISON BETWEEN SIMULATIONS (s ) AND (11) FOR L = 1

A= and

N 2

+ 0

e e[

e ]

d,

(14)

N 64 128 256 512 1024 2048 4096

2 s

Eq.(11) 1.64 1.64 1.64 1.64 1.64 1.64 1.64

1.36 1.46 1.52 1.57 1.6 1.63 1.6

B=N

+ 0

e e[

e ]

d.

(15)

Let us investigate the rst term A. A simple integration by parts yields mc (N ) = and
+ 0

(1 e[

e ]

)d,

(16)

TABLE III M EAN VALUES COMPARISON BETWEEN SIMULATIONS (ms ) AND (10) FOR L=4 N 64 128 256 512 1024 2048 4096 ms 5.58 6.31 7.03 7.75 8.47 9.18 9.91 Eq.(10) 4.73 5.42 6.12 6.81 7.5 8.2 8.89

dmc (N ) = dN

+ 0

e e[

e ]

d.

(17)

A is then exactly equal to N dmc (N ) . (18) 2 dN On the other hand, exact calculation of B happens to be really difcult. Plot of factor B shows that it can be approximated by something like ln(N ); meaning A= B ln(N ). (19)

TABLE IV
2 VARIANCE VALUES COMPARISON BETWEEN SIMULATIONS (s ) AND (11)

FOR

L=4 Eq.(11) 1.64 1.64 1.64 1.64 1.64 1.64 1.64

Here natural logarithmic function is used for approximation as this function can equate factor B using minimum number of coefcients. Finally, mc is the solution of the following differential equation mc (N ) + from which we get mc (N ) (lnN + 1 + ), N2 = ln . e (21) N dmc (N ) ln(N ), 2 dN (20)

N 64 128 256 512 1024 2048 4096

2 s

1.52 1.62 1.69 1.72 1.74 1.71 1.76

As this mc (N ) is obtained analytically, thus will be labeled as mc,a (N ) to distinguish from simulation based formulations.

1683

2) Discussions: As N is quite large in practical OFDM transmissions, thus (1/N 2 ) can be neglected without loss of generality and Eq.(21) can be rewritten as mc,a (N ) (lnN + ), = ln . e (22)

In [1], the authors have proposed an empirical relationship of the PMEPR cumulative distribution function for continuous signals. This approximation is Fc,s () (1 e )1 N , (23)

where 1 2.8 is obtained with simulations without theoretical justications. Thus Fc,s () is a simulation based distribution function for continuous signals. Starting from this relation and analogy with Eq.(10), the mean PMEPR value is about mc,s (N ) lnN + + ln1 lnN + 1.60. (24)

Fig. 2. Comparison between Exact and Approximate Analytical Mean PMEPR Values

A. Mean PMEPR After implying Non-linear Least Square method, values of the coefcients and in Eq.(19) are found to be 1.07 and 5.12 respectively. Thus mean PMEPR would be 5.12 mc,a (N ) (lnN + ln ) = 1.07ln( N ). e e (26)

As mc,s (N ) lnN + 1.60, this conrms now that analytically the PMEPR cumulative distribution function for continuous signal can also be approximated under the form: Fc,a () (1 e )2 N ,

2 = ( ) e . e

(25)

Contrary to [1] and Eq.(23), an analytical form of the PMEPR cumulative function for continuous signals has been proposed with theoretical justications. IV. S IMULATIONS AND R ESULTS As mentioned before in previous section that Eq.(15) is almost unsolvable using basic mathematics. Thus the task of getting the approximate solution of the mentioned equation using minimum number of coefcients was achieved using curve tting technique. Initially this equation was plotted for different values of N to get the idea of its solutions form. Thus Eq.(15) plot revealed that it follows natural logarithmic curve shape. Thus c1 ln(c2 N ) was thought to be the solution. To get the values of the coefcients, Non-linear Least Square method was used. It is a type of Least Square method in which the coefcient values are obtained by minimizing the sum of the square of the difference between data value and tted response value. If yi and yi are ith data value and tted value after using estimated coefcients respectively, then target N is to minimize S = i=1 (yi yi )2 for N point data tting. The process of nding optimized coefcients is iterative in nature. It starts with initial estimated values of the coefcients and corresponding value of S is calculated. Based on certain tting algorithm, coefcient values are adjusted to minimize S. These algorithms include Trust-region, Levenberg-Marquardt, Gauss-Newton etc. Trust-region algorithm [7] is used here because of its improved and efcient performance to solve non-linear problems than others. The optimized results were obtained for 500 iterations of the algorithm.

One can easily see that mean PMEPR for discrete signals (Eq.(10)) is less than mean PMEPR for continuous signals (Eq.(26)) as probability of seizing the signal peak is more in the later case. Fig.2 presents the comparison between the exact solution of Eq.(12) and its tted solution (Eq.(26)) versus different N values. It seems that tted solution almost matches the exact solution. Also in Fig. 3 mean PMEPR values for different values of N , obtained by simulations (ms ), using Eq.(24) (mc,s (N )) and Eq. (26) (mc,a (N )) are compared. The results show that mc,a (N ) and mc,s (N ) are quite similar thus theoretically providing the truthfulness of Eq.(23). Also mc,s (N ) seems to be a better t than mc,a (N ) to simulation results ms . Similarly PMEPR distribution function Eq.(25) becomes Fc,a () (1 e )2 N = (1 e )2 N
1.07

(27)

where 2 = ( 5.12 )1.07 e . Fig. 4 compares PMEPR Come plementary Cumulative Distribution Function (CCDF = 1 CDF ) for continuous signals. The curves are plotted using Eq.(8), Eq.(23) and Eq.(27) for N = 64, 256 and 1024. Again similarity in distribution functions states the trueness of simulation based formula.

V. C ONCLUSION Theoretical analysis of two already developed PMEPR distribution functions is carried out by calculating rst statistical order of these distributions. One of these two distributions is simulation based with no theoretical justication. This papers

1684

[7] Branch, M.A., T.F. Coleman, and Y. Li, A Subspace, Interior, and Conjugate Gradient Method for Large-Scale Bound-Constrained Minimization Problems, SIAM Journal on Scientic Computing, vol. 21, no. 1, pp. 1-23, 1999.

Fig. 3.

Comparison between Mean PMEPR Values for different N

Fig. 4.

CCDF Comparison for different N Values

provides theoretical justication of the correctness of this distribution by showing that rst statistical order of both simulation-based and theory-based distribution functions is almost same. New PMEPR distribution function is derived analytically which satises already developed formulations. R EFERENCES
[1] R. Van Nee, A. De Wild, Reducing the Peak to Average Power ratio of OFDM, Proceedings of IEEE VTC, pp. 2072-2076, May. 1998. [2] X. Zhou , J. Caffery, A new Distribution bound and reduction scheme for OFDM PMEPR, International Symposium on Wireless Personal Multimedia Communications, vol. 1, pp. 158-162, Oct. 2002. [3] H. Ochiai and H. Imai, On the distribution of the Peak to Average Power Ratio in OFDM Signals, IEEE Trans. Comm., vol. 49, pp. 282-289, Feb. 2001. [4] M. Sharif, M. Gharavi-Alkhansari, and B. H. Khalaj, On the peak-toaverage power of OFDM signals based on oversampling, IEEE Trans. Comm., vol. 51, no. 1, pp. 72-78, Jan. 2003. [5] H. Ochiai and H. Imai, Peak-Power Reduction schemes in OFDM systems : a review, International Symposium on Wireless Personal Multimedia Communications, Yokosuka Research Park, pp. 247-252, Nov. 1998. [6] Jacques Palicot, Yves Lou t, Power Ratio denitions and analysis in e single carrier modulation, EUSIPCO, Antalya, Turkey, Sept. 2005.

1685

Potrebbero piacerti anche