IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 53, NO.

8, AUGUST 2006 1741

Joint Clip and Quantization Effects Characterization

in OFDM Receivers
Davide Dardari, Member, IEEE

Abstract—An accurate method to analyze the joint effect of with nonlinear devices) is used to evaluate the quantization
clipping, quantization and thermal noise in the analog-to-digital noise spectra for some types of input signals assuming that no
conversion in high-speed orthogonal frequency division mul- overloading (no clipping) is present. Most of the results are
tiplexing (OFDM) receivers is presented. The model provides
information about the spectral properties of the distortion noise, valid only for uniform quantization.
which allows the analytical characterization of the signal-to-noise The problem of evaluating the clipping effects has been ad-
distortion ratio at the input of the demodulator. Oversampling, dressed since the 1940’s, during world war II, with applications
filtering, uniform, and optimal nonuniform quantization effects to jamming of radar and communications systems [8]. Several
are taken into account. The optimal automatic gain control studies were carried out during successive decades. Recently,
working point is evaluated for each parameter configuration.
The differences from the results obtained by the classical pseu- this problem has became relevant due to the introduction of mul-
doquantization noise model are highlighted showing that, due to ticarrier modulations such as OFDM, which are characterized
the spectral characteristics, the gain introduced by oversampling by a high PAPR, and hence have to be subjected to clipping
is generally a few decibels lower than what foreseen by the clas- [9]–[11]. In [11] the asymptotic distortion spectrum is evalu-
sical theory. Moreover, the impact of these effects on the OFDM ated in the presence of different clipping strategies, whereas in
receiver performance degradation is evaluated and the validity of
simplified approximations is discussed. [12]–[15] different strategies are presented to reduce and eval-
uate the PAPR effects in OFDM systems.
Index Terms—Quantization, digital-to-analog (D/A) conversion,
Only partial analysis of the joint distortion-quantization plus
digital communication, nonlinear distortion, overflow.
thermal noise spectra are present in the literature to the author’s
knowledge [16]–[18]. In this paper the clipping-quantization
I. INTRODUCTION theory is revised by means of an exact joint analysis in the case
where a Gaussian input signal with arbitrary spectrum shape is
present. Theoretical results are applied to the evaluation of the
R ECENT standards for high-speed wireless systems (e.g.,
wireless LAN) are based on orthogonal frequency division
multiplexing (OFDM) schemes and are operating at very high
performance loss in OFDM receivers. In particular, an analytical
characterization of the spectrum is carried out taking oversam-
bit rate [1]. At the receiver, the analog-to-digital (A/D) conver- pling, filtering, thermal noise, uniform and optimal nonuniform
sion is one of the most critical stages where both quantization quantization into account, that leads to a simple expression of
and saturation (clipping) effects have to be carefully considered the actual SDQNR at the input of the demodulator.
since OFDM signals are characterized by a high peak-to-av- The model presented allows the identification of the condi-
erage power ratio (PAPR) [2]. In [3], [4] it has been shown tions for which the PQN model is applicable and it is general
that, in order to evaluate the receiver performance, the distor- enough to be applied in every case where the input can be mod-
tion noise needs to be properly characterized from the statistical elled as a Gaussian process (e.g., OFDM, code division multiple
and spectral point of view. This is also required to optimize the access schemes [3], [19]). The optimum working point (i.e., the
working point of the automatic gain control (AGC) unit that pre- clipping ratio) of the AGC stage, inserted before the A/D con-
cedes the A/D stage. version, is evaluated both for the uniform and nonuniform quan-
In the literature, the clipping and the quantization effects tization laws. Moreover, the gain introduced by oversampling
are often addressed separately. The evaluation of quantization is analyzed and compared with the gain predicted by the PQN
effects is a classical subject for the signal processing theory. model. Finally, the performance loss at the OFDM receiver is
The problem is generally addressed considering the quanti- evaluated comparing two different approximated models for bit
zation error as an additive noise that is uniformly distributed, error probability computation.
uncorrelated with the input signal, and has a white spectrum.
This is often referred to as the pseudoquantization noise (PQN) II. A/D CONVERSION
model [5]. The hypothesis for the validity of this model are
discussed in the good tutorial presented by Widrow et al. [6]. In Fig. 1, the reference receiver scheme is reported where
In [7], the characteristic function method (often used to deal the in-phase (I) and in-quadrature (Q) components are extracted
from the received intermediate frequency (IF) bandpass OFDM
Manuscript received April 27, 2004; revised January 31, 2005, and September signal (with bandwidth ) and applied to a couple of A/D con-
2, 2005. This work was supported by MIUR, Italy. This paper was recommended verters (ADCs). Since the two components are considered to be
by Associate Editor W. Namgoong. independent and to have the same statistical characterization,
The author is with the WiLAB IEIIT-BO/CNR, CNIT, DEIS-University of
Bologna, 40136 Bologna, Italy (e-mail: [email protected]). we consider, without loss of generality, only the in-phase com-
Digital Object Identifier 10.1109/TCSI.2006.875170 ponent. The received baseband signal is composed of the
1057-7122/$20.00 © 2006 IEEE
1742 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 53, NO. 8, AUGUST 2006

Fig. 1. A/D conversion section.

useful signal (the baseband OFDM in-phase signal com- where and represent the inverse Fourier transform
ponent) plus the (independent) additive white Gaussian noise for continuous time and discrete time functions, respectively.
. Due to IF filtering, both components are supposed to be The quantization law is generically described by a staircase
band-limited in the bandwidth . In [3] it has been nonlinear function . The output is limited in the range
demonstrated that, for common OFDM system parameters, the (clipping), thus takes both clipping and quan-
OFDM signal can be well approximated as a band-lim- tization into account. The ratio is called clipping
ited Gaussian zero mean stationary process. Thus, the received ratio. We consider without loss of generality. In the
signal is band-limited Gaussian as well. case where quantization levels are depicted (
The ADC input power set by the AGC device is bits), the quantization-clipping function can be expressed as a
being the statistical expectation. The generic symmetric staircase function in the following way:
signal-to-noise ratio (SNR) before the A/D stage is

(4)
(1)

where , and . where represents the th quantization threshold value and

The purpose of the filter is to reduce the out-of-band compo- is the amplitude of the th quantization interval. The function
nents that arise after the quantizer in the case where oversam- is the rectangular function defined as follows:
pling is adopted. Only the A/D conversion effects are considered
in the paper, hence the use of floating point arithmetic is as-
sumed in the rest of the signal chain. The input power spectrum (5)
otherwise
can be arbitrary even though we will con-
sider only the rectangular shape, typically used to describe the In the case of midraiser uniform quantization [5] with step size
OFDM signals (the thermal component spectrum is supposed , we have
to be rectangular shaped as well assuming ideal bandpass fil-
tering in the IF stage). The function is the normalized
(total power set to 1) input power spectrum. Without loss of gen-
erality, we consider a normalized bandwidth, , for the input
signal set to 1. (6)
The signal is sampled at the sampling frequency In Fig. 2, the midraiser and midtread relationships are shown.
, where is the oversampling factor. The nor-
malized (with respect to ) spectrum of the sampled version III. CHARACTERIZATION OF THE QUANTIZED SIGNAL
of is
The quantizer stage can be viewed as a nonlinear device sub-
jected to a zero mean Gaussian input. Recalling the classical
Bussgang’s theorem [23], the output signal can be decom-
(2)
posed in a nondistorted component, proportional to the input
signal , and a distortion component, , as here reported
Let us consider the generic samples and
. The normalized autocorrelation function is
(7)

The distortion component takes the distortion noise due to

(3) the nonlinearity into account (in this case clipping and quanti-
DARDARI: JOINT CLIP AND QUANTIZATION EFFECTS CHARACTERIZATION 1743

limiter) and (soft limiter), it easy to verify that (10) be-

comes, respectively

(12)
which is well known as the arcsin law [8], and

(13)

already derived in [8] and [9] following different approaches.

Starting from (10), applying the Fourier transform, we obtain
Fig. 2. Example of midraiser and midtread quantization laws. an expression for the spectrum of

zation), whereas the coefficient can be calculated as follows

[23]:

(8)

(14)
The component is, in general, nonGaussian and it is not
correlated to . No further information is available from the where we have denoted as the -fold convolution of
Bussgang’s Theorem about the power spectrum of the distortion with itself.
noise. In [3], [4] it has been shown that the performance of a In the following, as already anticipated, we will refer
OFDM scheme is strictly related to the power of the undistorted to rectangular shaped spectrums for , i.e.,
component, which is proportional to , and the power spectral . In this case, a closed form for exists
characteristic of the filtered distortion noise . [21]
The expressions of the power spectrum of and its filtered
version at the output of the filter, are now derived. The first
step is the evaluation of the autocorrelation function related to
the samples at the output of the quantization device, that is (15)

(9) for and zero otherwise. The symbol

returns the greatest integer . This expression can be well
In the Appendix the following expression for the autocorrelation
approximated for with the Gaussian function
function is carried out:

(16)
(10)
having defined . Finally, we can derive the expres-
where the coefficients are defined as sion of

(17)

The kind of series expansion in (14) is typical when Gaussian

(11)
signals are processed by nonlinear devices [8]. Here a closed
form expression of the coefficients has been derived for joint
The function is the Hermite polynomial of order [24] clipping and quantization nonlinearity. This result is also in
and the dependence on the clipping ratio has been made ex- agreement with the Bussgang’s Theorem; in fact, the output
plicit. In the limit cases where the number of levels (hard spectrum is composed by a undistorted component (first term of
1744 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 53, NO. 8, AUGUST 2006

Fig. 3. Normalized PSD (in decibels) as a function of the normalized fre- Fig. 4. SDQNR (in decibels) at the output of the quantizer stage versus clipping
quency, ADC resolution b = 4. (a)  = 0 dB. (b)  = 6 dB. (c)  = 10 dB. ratio  (in decibels), for different number of quantization bits (lines refer to the
(d) Undistorted component. PQN model). SNR !1 .

(14)) and a sum of distorted spectrums. The undistorted compo- It represents the ratio between the undistorted useful compo-
nent of the output signal has power equal to nent power and the total noise power (i.e., the undistorted and
distorted thermal noise plus the distorted useful signal compo-
nent). The quantity
(18)

The spectrum of the distortion component is the second (21)

term of (14). Recalling the result of the Bussgang’s theorem in
considers the fact that, due to the presence of the undistorted
(8), it is easy to verify that .
thermal noise component , not all the power of is useful
In Fig. 3, the power-spectrum density (PSD), of the distortion
for the demodulator.
component is compared with the spectra of the undistorted
component (curve d) for different clipping ratio and oversam- A. Uniform Quantization
pling factors. Curves are normalized to the useful output power
The behavior of SDQNR as a function of for different
and the sampling frequency. The ADC resolution is . As
values of ADC resolution, , is shown in Fig. 4 for
can be noted, when increases (far from the saturation) and
(thermal noise free case). It is compared with the classical
is small, the distortion spectrum tends to become white (i.e.,
signal-to-quantization noise ratio (SQNR) valid under the PQN
the PQN hypothesis become valid). In OFDM systems it is im-
model hypothesis [7], . For each value
portant to estimate the effect of the distortion inside the useful
of , there is an optimum value for the clipping ratio which
bandwidth; in fact, if the spectra is far from being white, each
optimizes the SDQNR, i.e., the working point for the AGC. To
sub-carrier is differently affected by the distortion noise. For ex-
the right of this point the SDQNR is limited by the granular
ample, looking at curve (a), a difference of 3 dB is present be-
noise, then the curves match those obtained through the PQN
tween the center and border frequency distortion components.
model that consider only this effect. To the left of the optimum
This aspect will be addressed in Section V.
point, the clipping phenomena becomes dominant. From these
IV. SDQNR EVALUATION results, the accuracy range of the PQN model for the evaluation
of the SQNR can be deduced. Also for high values of and
The total power at the output of the quantizer device is
low number of bits the PQN model fails (the quantization noise
becomes no more uniform in the step interval). The optimum
point depends slightly on and it is confined, for the considered
values of , in the range 6–11 dB.
In Fig. 5, the SDQNR is illustrated for two different values of
(19) SNR, dB (curves b-e) and dB (curves c-f)
and compared to the noise free case (curves a-d). Two values
where is the complementary error function. We can
for the number of quantization bits are reported, (curves
now introduce the signal-to-distortion-quantization noise ratio
a-b-c) and (curves d-e-f). As can be noted, the impact
(SDQNR) defined at the output of the quantizer stage
of the thermal noise contribution becomes significant when the
A/D converter is working far from the saturation point (large
values of ). For each clipping factor, if we decrease the SNR,
(20)
in order to keep the clipping factor constant (fixed AGC working
DARDARI: JOINT CLIP AND QUANTIZATION EFFECTS CHARACTERIZATION 1745

Fig. 5. SDQNR (in decibels) at the output of the quantizer stage versus clipping Fig. 6. SDQNR (in decibels) at the output of the quantizer stage versus clipping
ratio  (decibels), for different values of SNRs. b = 4: curves a-b-c, b = 6: ratio (in decibels). ADC resolution b = 4. Comparison between uniform and
curves d-e-f. Curves a-d: SNR !1 , curves b-e: SNR = 10 dB, curves c-f: optimal Lloyd–Max nonuniform quantization. SNR !1 .
SNR = 5 dB.

TABLE I
COEFFICIENTS B FOR DIFFERENT VALUES OF F . RECTANGULAR
point), due to the presence of the thermal noise component the SPECTRUM SHAPE
useful component will be decreased accordingly. This effect,
added to the distortion components of both useful and noise
signals, makes the SDQNR degradation significant even in the
presence of relatively high SNR (e.g., dB).

B. Nonuniform Quantization
The presence of the AGC stage (that keeps the statistical char-
acteristics of the input signal constant) and the Gaussian approx-
imation about the statistics of the input signal, enables the op-
timization of the quantizer levels in a Lloyd–Max sense [20]
leading to a nonuniform quantization. The gain introduced by Obviously, the undistorted component remains still undistorted
the optimization is pointed out in Fig. 6 in the case of 16 quan- after the filter processing phase, hence . Coefficients
tization levels. A maximum gain of about 2 dB is obtained far can be pre-calculated once the input spectrum shape and
from the saturation region. This gain reduces as the saturation the oversampling factor are known. They do not depend
region is approached. At the optimum working point the gain is on and on the quantizer device characteristic. In the case of
reduced by 1 dB. If the nonuniform quantizer is implemented rectangular spectrum shape, using (16), the coefficient with
by a compander, a PAPR reduction benefit can be also achieved can be well approximated with
by putting the compressor stage at the transmitter, as proposed
in [22] in the case of the A-law companding scheme.

C. Oversampling Gain Evaluation (24)

After the filter (ideal low-pass filter with cut frequency
In Table I the coefficients are reported for different values of
), the total distortion power, associated to signal , can
in the rectangular spectrum shape case. As can be noted, as
be evaluated by integrating the power spectrum at the output of
increases, in the limit the value of tends towards the ratio
the quantizer in the useful signal band
. Changing the input signal spectrum shape means only
calculating coefficients using (23).
Once is known, we can evaluate the SDQNR at the
output of the filter (demodulator input)
(22)

where (25)

It is expected to be greater than due to the filtering

effect of the distortion spectrum components that fall outside the
(23) useful band. In order to emphasize the role of spectrum shape
1746 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 53, NO. 8, AUGUST 2006

Considering ideal carrier recovering and time synchroniza-

tion, the bit error probability is

(28)

We name this model multivalue model. The last equation as-

sumes that all noise contributions are Gaussian and indepen-
dent.1 A simpler way to evaluate the bit error probability con-
sists in assuming the SDQNR be constant for each subcarrier
(single-value model) and equal to that evaluated in (25). In this
case, the bit error probability expression becomes simpler

(29)
Fig. 7. Oversampling gain factor,  (in decibels), versus clipping ratio (in deci-
bels), for different values of oversampling values. (a) F = 2, (b) (d) (e)
F = 4, (c) F = 8. Curves refer to SNR !1 exception made for curves In Fig. 8, the bit error probability as a function of SNR is re-
(d) and (e) which are for SNR = 20 and 10 dB, respectively. ADC resolution ported for two different values of A/D operating point (curves
b = 4.
b-d-f-h: dB, curves a-c-e-g: dB). Results ob-
tained using (28), curves a-e-f-h, and (29), curves c-g-b-d, are
and filtering on the SDQNR, it is convenient to introduce the compared for two different values of the oversampling factor
oversampling gain factor (curves a-b-c-d: , curves e-f-g-h: ). For low
values of , the performance is limited by the distortion noise
leading to a significant performance loss. When the distortion
(26) noise is not negligible, the approximated expression (29), based
on the single SDQNR value model, gives optimistic results.
that gives a measure of the benefits introduced by the oversam- In this case, to obtain accurate performance evaluations, the
pling process taking the actual spectral characteristic into ac- spectral characteristic of the distortion noise must be accounted
count. through (28). On the contrary, when the PQN model is valid
In Fig. 7, the oversampling gain factor is plotted against the (e.g., dB), no difference exists between the two models.
clipping ratio for and ADC resolu- In fact, in this case the distortion spectrum is practically white
tion . For all three cases, the maximum gain is equal to , and the performance evaluated using (28) (that takes the spec-
as foreseen by the PQN model but, especially for higher values trum shape into account) and (29) (that considers it constant)
of , the colored characteristics of the distortion spectrum be- are the same. Comparing curves and it is possible to have an
comes more relevant. In this case the gain is reduced by several idea of the gain obtained by oversampling. These results have
decibels and it is possible to identify the range of where the been confirmed by simulation (dotted points) thus validating the
PQN model is valid. The effect of SNR on the oversampling Gaussian hypothesis as done in [3], [4] in the case of high power
gain is reported in curves d) and e) which have been obtained amplifier nonlinearities. The analysis presented holds also in the
for and 10 dB, respectively. As can be noted, the presence of higher order constellation sizes provided that the
effect of thermal noise reduces the impact of the oversampling. Gaussian hypothesis is verified (some considerations about this
are present in [3] and [4]) and equations (28)–(29) are changed
V. OFDM RECEIVER PERFORMANCE accordingly. In this case, the AGC working point does not de-
pend on the signal structure.
In order to investigate the impact of nonlinear distortion on
the receiver performance due to quantization, saturation and
thermal noise, we consider an OFDM scheme employing VI. CONCLUSION
subcarriers and 4-QAM signaling scheme (guard interval and In this paper, the quantization, clipping, and thermal noise
pilot subcarriers are neglected for convenience without loss of effects in OFDM receivers have been jointly analyzed taking
generality). In this case the frequency spacing between subcar- different aspects like oversampling, filtering, uniform and op-
riers is . timum nonuniform quantization into account. The model devel-
The SDQNR at the th FFT block output, that depends on the oped is quite general and it is valid for any shape of the input
distorted component spectral characteristic according to (14), is spectrum and for any kind of quantizer law (e.g., uniform or
not uniform). The SDQNR and the optimal working point for
the AGC stage has been evaluated for different configurations.
1Actually, nonlinear noise due quantization is slightly correlated among sub-
(27)
cariers (the spectrum is not perfectly white as previously shown). However, sim-
where functions and have been considered ulation results reported in Fig. 8 will show that this assumption leads to accurate
constant in each frequency interval . bit error probability evaluations.
DARDARI: JOINT CLIP AND QUANTIZATION EFFECTS CHARACTERIZATION 1747

leading to

(33)
The following series expansion for holds [24]:

(34)

with
(35)
Fig. 8. Bit error probability versus signal-to-noise ratio. ADC resolution b = 4.
Comparison between two approximated models.  = 0 dB: curves a-e-c-g;
 = 10 dB: curves b-d-f-h. Single value model: curves c-g-b-d. Curves a-b-c-d: where
F = 1, curves e-f-g-h: F = 2. Circles refer to simulation results.
(36)

It has been verified that the common PQN model is not suf- is the one-dimensional Gaussian Q-function, i.e., the comple-
ficient to describe the A/D conversion process when the input ment of the cumulative distribution function corresponding to
signal is characterized by a high PAPR, as happens in OFDM the normalized (zero mean, unit variance) Gaussian random
receivers. Moreover, the oversampling gain in terms of SDQNR variable. The function is the Hermite polynomial of
is reduced by some dBs if the clipping effect is taken into ac- order [24]. By substituting this result in (33) we can write
count. Finally, the OFDM receiver performance degradation has
been evaluated comparing two different approximated models
for the bit error rate computation. Results show that the simpli-
fied model based on the single SDQNR value leads to optimistic
result when the ADC working point is located outside the PQN
model validity region.

APPENDIX I
Starting from (4) and (9) the autocorrelation function of (37)
can be written as Resembling terms in (37), we get

(30)
(38)
Denoting with the joint-Gaussian probability
density function (pdf) we have
Finally, we obtain (10) and (11). Since is in general an
anti-symmetric function, only the terms in (38) for odd are
different from zero.

REFERENCES
(31) [1] Wireless LAN Medium Access Control (MAC) and Phyhsical Layer
(PHY) Specifications, High-Speed Physical Layer in the 5 GHz Band,
Equation (31) can be simplified by introducing the normal- IEEE802.11a Standard. Part 11, 1999.
[2] D. J. G. Mestagh, “Calculation of ADC resolution for DMT modula-
ized two-dimensional Gaussian probability integral [24] tion,” Electron. Lett. , vol. 31, no. 16, pp. 1315–1316, Aug. 1995.
[3] D. Dardari, V. Tralli, and A. Vaccari, “A theoretical characterization of
nonlinear distortion effects in OFDM systems,” IEEE Trans. Commun.,
vol. 48, no. 10, pp. 1755–1764, Oct. 2000.
[4] P. Banelli and S. Cacopardi, “Theoretical analysis and performance of
(32) OFDM signals in nonlinear AWGN channels,” IEEE Trans. Commun.,
vol. 48, no. 3, pp. 430–441, Mar. 2000.
1748 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 53, NO. 8, AUGUST 2006

[5] L. R. Rabiner and R. W. Schafer, Digital Processing of Speech Sig- [22] X. Wang, T. T. Tjhung, C. S. Ng, and A. A. Kassim, “On the
nals. Englewood Cliffs, NJ: Prentice-Hall, 1978. SER analysis of A-law companded OFDM system,” Proc. IEEE
[6] B. Widrow, I. Kollar, and M. C. Liu, “Statistical theory of quantization,” GLOBECOM’00, vol. 2, pp. 756–760, 2000.
IEEE Trans. Instrum. Measur., vol. 45, no. 2, pp. 353–361, Apr. 1996. [23] H. E. Rowe, “Memoryless nonlinearities with gaussian inputs: Elemen-
[7] R. M. Gray, “Quantization noise spectra,” IEEE Trans. Inf. Theory, vol. tary results,” Bell Syst. Tech. J., vol. 61, no. 7, pp. 1519–1525, Sep.
36, no. 6, pp. 1220–1244, Nov. 1990. 1982.
[8] J. H. Van Vlek and D. Middleton, “The spectrum of clipped noise,” [24] M. Abramowiz and I. E. Stegun, Handbook of Mathematical Func-
Proc. IEEE, vol. 54, no. 1, pp. 2–19, Jan. 1943. tions. Washington, DC: National Bureau of Standards, 1972.
[9] R. Gross and D. Veeneman, “SNR and spectral properties for a clipped
DMT ADSL signal,” in Proc. IEEE Int. Conf. Commun. (ICC’94), vol.
2, pp. 843–847.
[10] D. J. G. Mestagh, P. M. P. Spruyt, and B. Biran, “Analysis of clip- Davide Dardari received the Laurea degree in elec-
ping effect in DMT-based ADSL systems,” in Proc. IEEE Int. Conf. tronic engineering (summa cum laude) and the Ph.D.
Commun. (ICC’94), vol. 1, pp. 293–300. degree in electronic engineering and computer sci-
[11] J. E. Mazo, “Asymptotic distortion spectrum of clipped, dc-biased, ence from the University of Bologna, Bologna, Italy,
gaussian noise,” IEEE Trans. Commun., vol. 40, no. 8, pp. 1339–1344, in 1993 and 1998, respectively.
Aug. 1992. In the same year, he joined the Dipartimento di
[12] K. Sathananthan and C. Tellambura, “Coding to reduce both PAR and Elettronica, Informatica e Sistemistica, University
PICR of an OFDM signal,” IEEE Commun. Lett., vol. 6, no. 8, pp. of Bologna, to develop his research activity in the
316–318, Aug. 2002. area of digital communications. He has collaborated
[13] H. Chen and G. J. Pottie, “An orthogonal projection-based approach and taken a significant role in the following main
for PAR reduction in OFDM,” IEEE Commun. Lett., vol. 6, no. 5, pp. National and European Projects: European project
169–171, May 2002. “PROMETHEUS” regarding short-range communication systems for coop-
[14] B. S. Krongold and D. L. Jones, “An active-set approach for OFDM erative driving; MIUR “WWLAN” project for wideband high-speed wireless
PAR reduction via tone reservation,” in IEEE Trans. Signal Process., LAN; CNIT/ASI (Italian Space Agency) Projects “Integration of Multimedia
Feb. 2004, vol. 52, no. 2, pp. 495–509. Services on Heterogeneous Satellite Networks,” “Study, Design, and Realiza-
[15] D. Dardari, V. Tralli, and A. Vaccari, “A novel low complexity tech- tion of Guaranteed Quality of Service Re-configurable Satellite Networkfor
nique to reduce nonlinear distortion effects in OFDM systems,” in Proc. Multimedia Applications” and WAVE; MIUR Projects “VICOM” (Virtual
9th IEEE Int. Sympo. Personal, Indoor and Mobile Radio Communica- Immersive Communications) and CRIMSON; European Project “Phoenix,”
tions (PIMRC’98), Boston, Sep. 1998, vol. 2, pp. 795–800. European Network of Excellence “NEWCOM.” From 2000 until 2005, he was
[16] D. S. Zrnic, “Signal-to-noise ratio of nonlinear devices,” IEEE Trans. a Research Associate at the University of Bologna. He also held the position
Inf. Theory, vol. IT-21, no. 6, pp. 662–663, Nov. 1975. of Lecturer and contract Professor of Electrical Communications and Digital
[17] W.J. Hurd, “Correlation function of quantized sine wave plus Gaussian Transmission and Telecommunications Systems at the same university. He is
noise,” IEEE Trans. Inf. Theory, vol. 13, no. 1, pp. 65–68, Jan. 1967. currently an Associate Professor at the University of Bologna at Cesena, Cesena
[18] W. B. Davenport Jr. and W. L. Root, An Introduction to the Theory of (FC), Italy. During the winter 2005, he was a Research Affiliate at Massa-
Random Signals and Noise. New York: Mc Graw Hill, 1958. chusetts Institute of Technology (MIT), Cambridge. His research interests are
[19] A. Conti, D. Dardari, and V. Tralli, “An analytical framework for in orthogonal frequency division multiplexing systems, ultrawide bandwidth
CDMA systems with nonlinear amplifier and AWGN,” IEEE Trans. communication and localization, wireless sensor networks, wideband wireless
Commun. , vol. 50, no. 7, pp. 1110–1120, Jul. 2002. LAN.
[20] J. Max, “Quantizing for minimum distortion,” IRE Trans. Inf. Theory, He serves as Editor for IEEE TRANSACTIONS ON WIRELESS
vol. IT-6, pp. 7–12, Mar. 1960. COMMUNICATIONS, and as TPC member for the Wireless Communications
[21] D. R. Morgan, “Finite limiting effects for a band-limited Gaussian Symposium at IEEE International Conference on Communications (ICC
random process with applications to A/D conversion,” IEEE Transa. 2004–ICC 2006). He is co-chair of the International Conference on
Acoust., Speech Signal Process., vol. 36, no. 7, pp. 1011–1016, Jul. Ultra-Wideband (ICUWB 2006) and ICC 2007 Wireless Communications
1988. Symposium.