Professional Documents
Culture Documents
Spooner2001 CyclicSpecAnlysContinuousPhaseModSigs
Spooner2001 CyclicSpecAnlysContinuousPhaseModSigs
1, JANUARY 2001
Abstract—Continuous-phase modulated (CPM) signals play a a new kind of CPM modulator. Reduced-complexity LR-based
prominent role in modern communication systems due to their Viterbi-algorithm demodulators are derived in [1], [5], [6], and
desirable constant-modulus property and the ability to control [21]. Work on related topics has also begun to appear. For ex-
their power and bandwidth efficiencies. Popular CPM signals
include the classical minimum-shift keyed (MSK) signal, the ample, in [9], the LR is used to extend linear-modulation blind-
LREC family of signals also known as continuous-phase fre- equalization techniques to GMSK.
quency-shift-keyed (CPFSK) signals, and Gaussian MSK, which The LR is not the only linear approximation to CPM signals.
is used in state-of-the-art GSM and PCS mobile communication A parallel line of inquiry has been developed around the notion
systems. CPM signals, like virtually all man-made communication of approximating a CPM signal with a single staggered quater-
signals, are known to exhibit cyclostationarity, which implies that
their probabilistic parameters, such as mean, second moment, and nary phase-shift keyed (SQPSK) signal. This idea is based on
higher order cumulants, are almost-periodic functions of time. In the observation that MSK is exactly represented as a CPM signal
this paper, a novel representation of CPM signals as a sum of PAM (the 1REC CPFSK signal with modulation index of )
signals is presented for both integer and noninteger modulation and as a SQPSK signal with half-cosine pulse functions. The
index cases. Then, the th-order cyclostationarity properties of approximation itself is considered in [25], in which the SQPSK
binary CPM signals are derived in terms of th-order temporal
and spectral moment and cumulant functions. Moreover, the pulse function is sought that minimizes the mean-square error
case of -ary CPM signals is briefly addressed. The results are between the CPM signal and the SQPSK signal. Finally, in [4]
illustrated with simulations involving MSK, REC, and GMSK and [8], the SQPSK representation is used to develop reduced-
signals. complexity Viterbi algorithms for demodulation of GMSK sig-
Index Terms—Continuous-phase modulated signals, cyclosta- nals.
tionarity, higher order statistics. In this paper, the Laurent representation of a binary CPM
signal with noninteger modulation index in terms of PAM sig-
nals is revised. Specifically, with reference to CPM signals with
I. INTRODUCTION
finite-length phase-response functions, a general procedure is
[17], [32] and are typically simple linear functions of funda- is referred to as the phase response of the system. The pulse am-
mental modulation parameters such as keying rates and carrier plitude is normalized so that , where is referred
frequencies. to as the modulation index.
Examples of exploitation of second-order cyclostation-
arity for detection and modulation classification are given in A. PAM Expansion for Noninteger Modulation Index
[14]–[16], and [22], in which knowledge of cycle frequencies For , the CPM signal can be represented
is used to compute decision statistics. More recently, higher as
order cyclostationarity [17], [30], [32] has been applied to the
modulation classification problem [24], [26], [33]. The various
proposed algorithms require, in one way or another, detailed
knowledge of the higher order moments and cumulants of the
(3)
cyclostationary signals of interest.
Moreover, for noninteger values of the modulation index , it
Since CPM signals are assuming an increasing prominence
can be easily shown that
in modern communication systems and high-performance
signal processing algorithms of various sorts benefit from de- (4)
tailed knowledge about communication signals’ time-varying
moments and cumulants, we present, in this paper, a derivation where
of the th-order cumulants of binary CPM signals. Such a
problem was addressed in [10] and [11], where it was limited to (5)
the fourth-order case and with reference to full-response CPM
signals. Moreover, second-order cyclic spectra of some CPM
signals were derived in [18], [35], and [36]. The analysis made (6)
in the present paper is based on an expansion of binary CPM
signals in terms of PAM signals, which is a revised version of and hence
that presented in [23] and facilitates the subsequent computa-
tion of the desired probabilistic parameters. Both integer and
noninteger modulation index cases are considered. Moreover,
starting from a result presented in [27], the case of -ary CPM
signals is briefly addressed. The results are illustrated using
CPFSK, MSK, and GMSK; signals and simulations are used to
corroborate the theoretical analysis.
The paper is organized as follows. The derivation of the PAM
signal representation of CPM signals is sketched in Section II.
The representation is used in Section III to derive the th-order T
rect (8)
(1)
where
where is a sequence of binary independent identically
rect (9)
distributed (i.i.d.) equiprobable symbols assuming values , otherwise
is the bit rate, is a finite-length pulse nonzero only for
, and it follows that the CPM signal (1) can be written as
(10)
(2)
32 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001
T
(11)
and even
(21)
rect (12) odd
where is an undetermined constant that can assume values
That is, the CPM signal can be expressed as sum of PAM . Therefore, the CPM signal (1) can be written as
signals with baud rate , base pulses with width , and
statistically dependent modulating sequences . (22)
Representation (10) can be reduced to the sum of only
PAM signals. For this purpose, it is shown in Appendix A that
all vectors can be grouped into classes , where
each constituted by the elements , and it T
even
results that T (23)
odd
and
(13)
rect (24)
where
Finally, let us note that in the case of integer modulation index
(14) the sum of PAM signals cannot be simplified to a sum of
signals, as in the case of noninteger modulation index.
with
III. HIGHER ORDER CYCLOSTATIONARITY PROPERTIES OF
T
(15) CPM SIGNALS
and In this section, higher order cyclic parameters [17] are de-
rived for a CPM signal, starting from its representation as sum
(16) of PAM signals. Moreover, the gross higher order cyclostation-
arity properties of the signal are discussed.
Note that the procedure to reduce the number of PAM signals A. Background on Higher Order Cyclostationarity
from to is different from that (not completely speci-
fied) in [23] and leads to PAM signals whose pulses (16) are In the fraction-of-time (FOT) probability framework, con-
expressed, unlike in [23], in terms of the nonoverlapping pulses tinuous-time complex-valued time-series are
. said to exhibit joint th-order wide-sense cyclostationarity with
cycle frequency , for a given conjugation configura-
B. PAM Expansion for Integer Modulation Index tion, if the th-order cyclic temporal cross-moment function
(CTCMF)
Since the complex exponential function is easily expressed in
terms of elementary trigonometric functions
(25)
(17)
where T T T , and T.
T The function , which is referred to as the th-order re-
duced-dimension CSCMF (RD-CSCMF), can be expressed as
the -fold Fourier transform of the th-order reduced-di-
(20) mension CTCMF defined by setting into (25).
NAPOLITANO AND SPOONER: CYCLIC SPECTRAL ANALYSIS OF CPM SIGNALS 33
(32)
cum
(27) where are the Fourier transforms of the real-valued pulses
of the Fourier-series expansion of the th-order temporal defined in (16), and and are
cross-cumulant function [17]. Its -fold Fourier transform is the th-order RD-CSCMF and th-order CCP, respectively,
the th-order cyclic spectral cross-cumulant function , of the discrete-time-series defined
which can be written as [17] according to (15). The RD-CSCMF is the
T -fold discrete Fourier transform of the reduced-dimen-
(28)
sion cyclic temporal cross moment function (see Appendix B)
where the th-order cyclic cross-polyspectrum (CCP)
is the -fold Fourier transform of the reduced-dimension
CTCCF obtained by setting into (27).
The CCP is the preferred spectral characterization of higher
order cyclostationarity because it, unlike the RD-CSCMF, is a
well-behaved function for time-series models of interest under
the mild assumption that samples of the time-series are asymp-
totically statistically independent as their separation increases T
without bound.
T
P
(31) T (35)
34 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001
where T, , , and
are defined according to (14)–(16), and
(36)
T
(41)
from which the cyclic autocorrelation function and the conju- for . Hence, from (48) and (49), it follows that
gate cyclic autocorrelation function follow immediately since,
for , it results that
(51)
(43)
T
or
(44)
or (52)
C. Cyclic Parameters for Integer Modulation Index where is the Fourier transform of the real-valued pulse
The th-order RD-CSMF of a CPM signal with integer mod- defined in (24), and is the th-order CCP
ulation index can be derived starting from (22), from which it of the discrete-time-series defined ac-
immediately follows that cording to (23). The CCP is the -fold
(45) discrete Fourier transform of the RD-CTCCF
that can be evaluated by a formula analogous to (35), where the
cyclic temporal cross moment functions of the discrete-time-se-
From (23) and accounting for (B.3), it follows that ries are determined according to
even
(46)
odd
even
odd
T
(47)
where T, denotes the almost-periodic
component extraction operation for continuous time-series, and,
in derivation of (47) for odd, the fact that is periodic
and, hence, deterministic in the FOT probability sense has been
accounted for. T
From (47), it follows that CPM signals with integer modula-
tion index contain first-order periodicities whose period is for
odd.
even and for odd. Therefore, (22) can be rewritten as
(53)
(48)
Note that in (53) and for even, and
where and for odd.
In the special case , is a PAM signal with modu-
(49) lating sequence
even
(54)
odd
and it results that
and pulse
cum
Therefore, it can be easily shown that is practically coincident with the complex envelope of the real
T band-pass CPM signal
even
T T (62)
odd, even
T T
provided that the carrier frequency is sufficiently greater than
the (theoretically infinite but practically finite) bandwidth of the
odd, odd signal (1) as it happens in all real situations. In such a case, the
(56) higher order cyclic spectra and the corresponding supports in the
space of , its analytic signal, and its in-phase and
and
T quadrature components can be related to those of the complex
even envelope (1) by exploiting the relationships derived in [19].
odd
(57) IV. -ARY CPM SIGNALS
In this section, the higher order cyclostationarity properties
where and are the th-order cyclic temporal of -ary CPM signals are briefly addressed.
moment function and the th-order cumulant, respectively, of Let us consider the -ary CPM signal (with even)
the sequence . In derivation of (57) for odd, the fact that
for i.i.d. sequences with even first-order probability (63)
distribution function has been used. Moreover, from (51), (52),
(56), and (57), it follows that we have (58), shown at the bottom
where is a sequence of -ary i.i.d. symbols assuming
of the page, and
values in , and is the pulse de-
fined in (2).
T
It is shown in [27] that the CPM signal (63) can be expressed
even as the product
odd
(64)
(59)
T (65)
(60)
T
even
T
odd, even (58)
T
odd, odd
NAPOLITANO AND SPOONER: CYCLIC SPECTRAL ANALYSIS OF CPM SIGNALS 37
where and denotes a joint continuous where for even and for odd, is the
-fold (in ) and discrete (in ) convolution, from (64), periodic function
it follows that the RD-CSMF of the -ary CPM signal can be
written as rect
(73)
with for even and for odd, and
(68) is the real PAM signal
(70) rect
rect (76)
and
rect
(71)
or, equivalently, modulating sequence and rect
pulse . In the special case , one obtains
the expressions for the modulating sequence and the pulse of a (77)
PAM signal equivalent to the MSK signal. Furthermore, from (15) one has
From the above, it follows that the cyclostationarity proper-
ties of the 1REC signal with noninteger modulation index can (78)
be exactly determined by considering those of the PAM signal
with modulating sequence and pulse . That is, (40) and
provides the exact expression for the RD-CSMF of the 1REC
signal with noninteger . (79)
For integer values of the modulation index , from (22)–(24)
and (46)–(49), it follows that the CPFSK signal can be expressed In order to compare the CPM signal with the PAM signal
as with modulating sequence and pulse , simu-
lation experiments have been carried out by discretizing both
(72) time and frequency with sampling increments and ,
38 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001
Fig. 1. Real parts of the 2REC signal with T = 8T and h = 0:5 (solid line) Fig. 3. Magnitude of the conjugate second-order cyclic spectrum estimates at
and of its approximating PAM signal s (t) (dotted line). = 1=2T of the 2REC signal with T = 8T and h = 0:5 (thin line) and of
its approximating PAM signal s (t) (thick line).
Fig. 2. Power spectral density function estimates of the 2REC signal with T =
8 T and h = 0:5 (thin line) and of its approximating PAM signal s (t) (thick Fig. 4. Magnitude of the second-order cyclic spectrum estimates at = 1=T
line). of the 2REC signal with T = 8T and h = 0:5 (thin line) and of its
approximating PAM signal s (t) (thick line).
(a) (b)
(c) (d)
Fig. 6. Power spectral density function estimates of the 2REC signal with T = 8T (thick line) compared with that of the secondary PAM signal s (t) (thin
line) for (a) h = 0:2, (b) h = 0:4, (c) h = 0:6 and (d) h = 0:8.
Finally, an experiment has been carried out to study the close- concentrated in the interval . Therefore, a GMSK
ness of the approximation of the 2REC signal by the only one signal with pulse turns out to be practically equivalent
PAM signal for various values of the modulation index to a CPM signal with finite-width pulse
. By considering the power spectral density (PSD) of the dif-
ference signal [that is, the PSD of the rect (82)
secondary PAM ], it results that the PAM signal turns
out to be a good approximation to the 2REC signal for a wide
Such a CPM signal is, in turn, equivalent to the sum of two PAM
range of values of the modulation index ( ), whereas it is
signals whose pulses, accounting for (5), (6), (12), and (16), are
not sufficient to closely approximate the CPM signal for values
given by
of close to unity (see Fig. 6).
rect
rect (83)
erfc erfc (80) and
rect (84)
where , (typically ), and
erfc is the complementary error function defined by
and whose modulating sequences are given by (78) and (79),
respectively, specialized for :
erfc (81)
(85)
From an (amplitude scaled) plot of as a function of
(see Fig. 7), it is evident that almost all of the energy of is (86)
40 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001
(87)
where [37]
(b)
Fig. 8. (a) Real parts of the GMSK signal with T = 8T (solid line) and of
its approximation in terms of 2 PAM signals (dotted line). (b) Real parts of the
GMSK signal with T = 8T (solid line) and of its approximating PAM signal
s (t) (dotted line).
erfc
erfc
APPENDIX A
In this Appendix, the proof of (13) is reported.
Let us group all vectors T into
classes according to the following rule: and belong to
the same class if and only if for some , it results
Fig. 10. Magnitude of the conjugate second-order cyclic spectrum estimates that
at = 1=2T of the GMSK signal with T = 8T (thin line) and of its
approximating PAM signal s (t) (thick line). (A.1)
where
(A.2)
.
Accounting for (A.2), a necessary condition is that the first
elements of must be 1, and the last elements of must
be 0. Furthermore, in each class, only one vector, say , can
have , and all the other vectors of the same class can be
obtained as shifted versions of , provided that the shift is per-
formed shifting 1s into and 0s being shifted out of . More-
over, observing that a vector can be shifted times (shifting
1s into and 0s being shifted out of ), obtaining a vector
Fig. 11. Magnitude of the second-order cyclic spectrum estimates at = 1=T belonging to the same class only if the number
of the GMSK signal with T = 8T (thin line) and of its approximating PAM
signal s (t) (thick line).
(A.3)
Fig. 12. Magnitude of the slice for f = 1=2T and f = 0 of the fourth-order
cyclic polyspectrum P (f ; f ; f ) estimates at = 1=T of the GMSK
signal with T = 8T and h = 0:5 (thin line) and of its approximating PAM
signal s (t) (thick line).
(A.4)
spectral cumulant functions for binary CPM signals. Moreover,
-ary CPM signals are briefly addressed. Since CPM signals From rules 1)–3), it follows that different classes
are cyclostationary, their temporal cumulant functions are pe- exist. Moreover, contains
riodic (or almost-periodic) as functions of time. As such, they elements, and ( ) contains
can be represented as Fourier series, the frequencies of which elements where with an odd number
are called cycle frequencies and the coefficients of which are (i.e., is the power of 2 in the prime factor decomposition
called cyclic cumulant functions. of ). Therefore , each class with odd contains only one
42 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001
TABLE I
CLASSES P , VECTORS b , SETS I , AND VALUES OF d(b) FOR THE VECTORS b IN THE CASE L = 4
element, and each class with even contains the elements APPENDIX B
such that In this Appendix, the proof of (33) and (38) is provided.
The th-order cyclic temporal cross moment function of the
discrete-time-series is given by
(A.5)
(B.1)
where, in the last equality, (A.2) and (15) have been accounted
for. Moreover, since the are i.i.d., one has
(A.6) T
and, hence [4] G. Benelli, A. Garzelli, and F. Salvi, “Simplified Viterbi processors for
the GSM pan-European cellular communication system,” IEEE Trans.
T Veh. Technol., vol. 43, pp. 870–877, Nov. 1994.
[5] G. Castellini, F. Conti, E. Del Re, and L. Pierucci, “A continuously adap-
tive MLSE receiver for mobile communications: Algorithm and perfor-
T mance,” IEEE Trans. Commun., vol. 45, pp. 80–88, Jan. 1997.
[6] G. Colavolpe and R. Raheli, “Reduced-complexity detection and phase
T
(B.4) synchronization of CPM signals,” IEEE Trans. Commun., vol. 45, pp.
1070–1079, Sept. 1997.
otherwise [7] A. N. D’Andrea, A. Ginesi, and U. Mengali, “Frequency detectors
for CPM signals,” IEEE Trans. Commun., vol. 43, pp. 1828–1837,
the expression (33) of can be easily derived from Feb./Mar./Apr. 1995.
(B.2). [8] R. D’Avella, L. Moreno, and M. Sant’Agostino, “An adaptive MLSE
The cyclic temporal moment function of the modulating se- receiver for TDMA digital mobile radio,” IEEE Trans. Commun., vol.
7, pp. 122–129, Jan. 1989.
quence [9] Z. Ding and G. Li, “Single-Channel blind equalization for GSM cellular
systems,” IEEE J. Select. Areas Commun., vol. 16, pp. 1493–1505, Oct.
1998.
(B.5) [10] J. R. Fonollosa and J. A. R. Fonollosa, “Estimation of the modulation
index of CPM signals using higher-order statistics,” in Proc. IEEE Int.
Conf. Acoust., Speech, Signal Process., Minneapolis, MN.
is given by [11] , “Analysis of CPM signals using higher-order statistics,” in Proc.
IEEE Conf. Military Commun., Oct. 1993.
[12] W. A. Gardner, Statistical Spectral Analysis: A Nonprobabilistic
Theory. Englewood Cliffs, NJ: Prentice-Hall, 1987.
[13] W. A. Gardner, Ed., Cyclostationarity in Communications and Signal
Processing. New York: IEEE, 1994.
[14] W. A. Gardner and C. M. Spooner, “Cyclic spectral analysis for signal
detection and modulation recognition,” in Proc. IEEE Conf. Military
T Commun., San Diego, CA, Oct. 1988, pp. 419–423.
[15] , “Signal interception: Performance advantages of cyclic-feature
detectors,” IEEE Trans. Commun., vol. 40, pp. 149–159, Jan. 1992.
[16] , “Detection and source location of weak cyclostationary signals:
T
Simplifications of the maximum-likelihood receiver,” IEEE Trans.
Commun., vol. 41, pp. 905–916, 1993.
[17] , “The cumulant theory of cyclostationary time-series—Part I:
Foundation,” IEEE Trans. Signal Processing, vol. 42, pp. 3387–3408,
Dec. 1994.
T
[18] P. Gournay and P. Viravau, “Corrélation spectrale théorique des modula-
(B.6) tions CPM—Partie I: Resultat analytique pour les modulations CPFSK
à 2 états (1REC),” Ann. Télécommun., vol. 53, no. 7–8, pp. 267–278,
1998.
Then, accounting for (B.3), (B.4), and the limits [19] L. Izzo and A. Napolitano, “Higher-order statistics for Rice’s represen-
tation of cyclostationary signals,” Signal Process., vol. 56, pp. 279–292,
(B.7) Feb. 1997.
[20] P. Jung, “Laurent’s representation of binary digital continuous phase
modulated signals with modulation index 1/2 revisited,” IEEE Trans.
and Commun., vol. 42, pp. 221–224, Feb./Mar./Apr. 1994.
[21] G. K. Kaleh, “Simple coherent receivers for partial response contin-
uous phase modulation,” IEEE J. Select. Areas Commun., vol. 7, pp.
1427–1436, 1989.
[22] Y. K. Kim and C. L. Weber, “Generalized single-cycle classifier with
(B.8) applications to SQPSK vs. 2 PSK,” in Proc. IEEE Conf. Military
Commun., 1989.
(38) easily follows. [23] P. A. Laurent, “Exact and approximate construction of digital
phase modulation by superposition of amplitude modulated
pulses (AMP),” IEEE Trans. Commun., vol. COMM-34, pp. 150–160,
ACKNOWLEDGMENT Feb. 1986.
The authors would like to acknowledge Prof. W. A. Gardner [24] C. Le Martret and D. Boiteau, “A general maximum likelihood classifier
for modulation classification,” in Proc. IX Eur. Signal Process. Conf.,
of the University of California at Davis for his support and guid- Rhodes, Greece, Sept. 1998.
ance of the first author while he held the Visiting Post-Doctorate [25] M. Luise and U. Mengali, “A new interpretation of the average matched
Research Associate position. Moreover, Dr. S.V. Schell is also filter for MSK-type receivers,” IEEE Trans. Commun., vol. 39, pp.
14–16, Jan. 1991.
acknowledged for his insightful comments. [26] P. March, C. Le Martret, and J.-L. Lacoume, “Modulation classification
based on a maximum-likelihood receiver in the cyclic-HOS do-
REFERENCES main,” in Proc. IX Euro. Signal Process. Conf., Rhodes, Greece, Sept.
M
1998.
[1] N. Al-Dhahir and G. Saulnier, “A high-performance reduced-com- [27] U. Mengali and M. Morelli, “Decomposition of -ary CPM signals into
plexity GMSK demodulator,” IEEE Trans. Commun., vol. 46, pp. PAM waveforms,” IEEE Trans. Inform. Theory, vol. 41, pp. 1265–1275,
1409–1412, Nov. 1998. Sept. 1995.
[2] T. Aulin and C. W. Sundberg, “Continuous phase modulation—Part I: [28] M. Morelli, U. Mengali, and G. M. Vitetta, “Joint phase and timing re-
Full response signaling,” IEEE Trans. Commun., vol. COMM-29, pp. covery with CPM signals,” IEEE Trans. Commun., vol. 45, pp. 867–876,
196–209, Mar. 1981. July 1997.
[3] T. Aulin, N. Rydbeck, and C. W. Sundberg, “Continuous phase modu- [29] K. Murota and K. Hirade, “GMSK modulation for digital mobile radio
lation—Part II: Partial response signaling,” IEEE Trans. Commun., vol. telephony,” IEEE Trans. Commun., vol. COMM-29, pp. 1044–1050,
COMM-29, pp. 210–225, Mar. 1981. July 1981.
44 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001
[30] A. Napolitano, “Cyclic higher-order statistics: Input/output relations for Chad M. Spooner (S’88–M’92) was born in Des
discrete- and continuous-time MIMO linear almost-periodically time- Moines, IA, on October 5, 1963. From 1981 to
variant systems,” Signal Process., vol. 42, pp. 147–166, Mar. 1995. 1984, he attended Santa Rosa Junior College, Santa
[31] B. Picinbono, “Polyspectra of ordered signals,” IEEE Trans. Inform. Rosa, CA, and received the A.S. degree (with high
Theory, vol. 45, pp. 2239–2252, Nov. 1999. honors) in 1984. He received the B.S. degree from
[32] C. M. Spooner and W. A. Gardner, “The cumulant theory of cyclosta- the University of California, Berkeley, in 1986 and
tionary time-series, Part II: Development and applications,” IEEE Trans. the M.S. and Ph.D. degrees from University of
Signal Processing, vol. 42, pp. 3409–3429, Dec. 1994. California, Davis (UC Davis), in 1988 and 1992, all
[33] C. M. Spooner, “Classification of cochannel communication signals in electrical engineering.
using cyclic cumulants,” in Proc. Twenty-Ninth Asilomar Conf. Signals, From 1994 to 1997, he was with Mission Research
Syst., Comput,, Pacific Grove, CA, 1995, pp. 531–536. Corporation, Monterey, CA, where he conducted re-
[34] C. W. Sundberg, “Continuous-phase modulation,” IEEE Commun. search on spectral correlation analysis and its applications. In 1994, he gave a
Mag., vol. 24, pp. 25–38, Apr. 1986. series of invited lectures on cumulant-based signal processing at the University
[35] P. Viravau and P. Gournay, “Corrélation spectrale théorique des modu- of Napoli Federico II. From 1997 to 1999, he was a Consultant with Innovative
lations CPM—Partie II: Méthode de calcul générale et analyse,” Ann. Algorithms, San Francisco, CA, for which he conducted research on automatic
Télécommun., vol. 53, no. 7–8, pp. 279–288, 1998. modulation classification and interference-tolerant signal-parameter estimation.
[36] D. Vučic and M. Obradović , “Spectral correlation evaluation of MSK He returned to Mission Research Corporation in 1999. His research interests in-
and offset QPSK modulation,” Signal Process., vol. 78, pp. 363–367, clude weak-signal detection and parameter estimation, automatic modulation
Nov. 1999. classification, the theory of the higher order statistics of cyclostationary signals,
[37] “The integrator,” http://www.integrals.com, powered by Mathematica. wavelet-based image classification, and the design of secure communication
systems.
Dr. Spooner received the 1992 Award for Excellence in Ph.D. Research from
the UC Davis Chapter of Sigma Xi and the 1995 UC Davis Allen G. Marr Prize
Antonio Napolitano (M’95) was born in Napoli, in the physical sciences and engineering for his doctoral work.
Italy, on February 7, 1964. He received the Dr. Eng.
degree (summa cum laude) in electronic engineering
in 1990 and the Ph.D. degree in electronic and
computer engineering in 1994, both from the
University of Napoli Federico II.
In 1994 and 1995, he was Appointed Professor of
Electrical Communications for the Diploma degree
at the University of Salerno, Salerno, Italy. Since
1995, he has been an Assistant Professor with the
Department of Electronic and Telecommunication
Engineering, University of Napoli Federico II, where, since 1999, he has been
an Appointed Professor of Radar Theory and Methods. From January to June
1997, he was with the Department of Electrical and Computer Engineering,
University of California, Davis, as Post-Doctoral Research Associate. His
research interests include statistical signal processing, system identification,
and the theory of the higher order statistics of nonstationary signals.
Dr. Napolitano received the Best Paper of the Year Award from the Euro-
pean Association for Signal Processing (EURASIP) in 1995 for his paper enti-
tled “Cyclic higher-order statistics: Input/output relations for discrete- and con-
tinuous-time MIMO linear almost-periodically time-variant systems” (Signal
Process., vol. 42, Mar. 1995).