Escolar Documentos
Profissional Documentos
Cultura Documentos
net/publication/3479904
CITATIONS READS
662 2,340
4 authors, including:
Wei Su
US Army
67 PUBLICATIONS 2,669 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Octavia A. Dobre on 08 October 2014.
Abstract: The automatic recognition of the modulation format of a detected signal, the inter-
mediate step between signal detection and demodulation, is a major task of an intelligent receiver,
with various civilian and military applications. Obviously, with no knowledge of the transmitted
data and many unknown parameters at the receiver, such as the signal power, carrier frequency
and phase offsets, timing information and so on, blind identification of the modulation is a difficult
task. This becomes even more challenging in real-world scenarios with multipath fading,
frequency-selective and time-varying channels. With this in mind, the authors provide a compre-
hensive survey of different modulation recognition techniques in a systematic way. A unified
notation is used to bring in together, under the same umbrella, the vast amount of results and
classifiers, developed for different modulations. The two general classes of automatic modulation
identification algorithms are discussed in detail, which rely on the likelihood function and features
of the received signal, respectively. The contributions of numerous articles are summarised in
compact forms. This helps the reader to see the main characteristics of each technique.
However, in many cases, the results reported in the literature have been obtained under different
conditions. So, we have also simulated some major techniques under the same conditions, which
allows a fair comparison among different methodologies. Furthermore, new problems that have
appeared as a result of emerging wireless technologies are outlined. Finally, open problems and
possible directions for future research are briefly discussed.
3 Likelihood-based approach to AMC where vi ¼ [vi†1 vi†2]† and, vi1 and vi2 are vectors of unknown
quantities modelled as unknown deterministics and r.v.s,
Within the LB framework, AMC is a multiple composite respectively. Usually, vi1 and vi2 consist of parameters and
hypothesis-testing problem. The idea behind the LB-AMC data symbols, respectively.
is that the probability density function (PDF) of the Note that ALRT requires a multidimensional integration,
observed waveform, conditioned on the embedded modu- whereas GLRT requires a multidimensional maximisation.
lated signal, contains all information for classification. The difficulty of performing a multidimensional integration
Depending on the model chosen for the unknown quantities, for a large number of unknown quantities and the need for
three LB-AMC techniques are proposed in the literature: knowing the prior PDFs may render the ALRT impractical.
average likelihood ratio test (ALRT) [1 – 13, 22], general- Alternatively, maximisation over the unknown data
ised likelihood ratio test (GLRT) [14, 19, 20] and hybrid symbols in GLRT can lead to the same value of the LF
likelihood ratio test (HLRT) [14 –18, 21, 22]. Quasi for nested signal constellations, for example BPSK and
ALRT [3 – 5 7, 9 – 13] and quasi HLRT [22] are also QPSK, 16-QAM and 64-QAM, [14, 96 Chap. 6], which in
proposed in the literature. turn yields incorrect classification. Averaging over the
unknown data symbols in HLRT, however, removes the
ALRT: This approach treats the unknown quantities as nested constellations problem of GLRT. Finally we empha-
random variables (r.v.s) with certain PDFs. So, the likeli- sise that the estimates of the unknown quantities, as
hood function (LF) under the hypothesis Hi , representative by-products of GLRT and HLRT, are of interest for data
of the ith modulation, i ¼ 1, . . . , Nmod , is given by demodulation.
ð In a two-hypothesis classification problem, the decision is
LðiÞ
A ½rðtÞ ¼ L½rðtÞjvi ; Hi pðvi jHi Þ dvi ð4Þ made according to
H1
Sills [1] ALRT BPSK, QPSK, 16QAM, V29, carrier phase u AWGN
32QAM, 64QAM
Wei and Mendel [2] ALRT 16QAM, V29 – AWGN
Kim and Polydoros [3, 4] quasi-ALRT BPSK, QPSK carrier phase u AWGN
Huang and Polydoros [5] quasi-ALRT UW, BPSK, QPSK, 8PSK, 16PSK carrier phase u and timing offset 1 AWGN
Sapiano and Martin [6] ALRT UW, BPSK, QPSK, 8PSK – AWGN
Long et al. [7] quasi-ALRT 16PSK, 16QAM, V29 carrier phase u AWGN
Hong and Ho [8] ALRT BPSK, QPSK signal level a AWGN
Beidas and Weber [9] ALRT quasi-ALRT 32FSK, 64FSK phase jitter ffkgKk¼1 AWGN
Beidas and Weber [10, 12] ALRT quasi-ALRT 32FSK, 64FSK phase jitter ffkgKk¼2 and timing AWGN
offset 1
Panagiotu et al. [14] GLRT HLRT 16PSK, 16QAM, V29 carrier phase u AWGN
Chugg et al. [15] HLRT BPSK, QPSK, OQPSK carrier phase u, signal power S and AWGN
PSD N0
Hong and Ho [16] HLRT BPSK, QPSK signal level a AWGN
Hong and Ho [17, 18] HLRT BPSK, QPSK angle of arrival w AWGN
Dobre et al. [21] HLRT BPSK, QPSK, 8PSK, 16PSK, channel amplitude a and phase q flat fading
16QAM, 64QAM
Abdi et al. [22] ALRT quasi-HLRT 16QAM, 32QAM, 64QAM channel amplitude a and phase w flat fading
These modulations have been used in the simulations in the original papers
where a ¼ 1 and w ¼ 0 when there is no fading. We define computed by averaging over the constellation points
the observed per-symbol SNR gs ¼ VST/N0 , in which corresponding to the ith modulation format. This is done
V ¼ 1 when there is no fading. Otherwise, V ¼ Efa2g is by substituting (5) into (4) (see (37), in the Appendix)
the average fading power, with a as the channel amplitude.
Under fading, S is set equal to one. Y
K pffiffiffi
LðiÞ
A ½rðtÞ ¼ EsðiÞ fexp½2a S N01 RefRðiÞ
k g
k
k¼1
3.1 ALRT-based algorithms
a2 STN01 jsðiÞ 2
k j g ð10Þ
In this section, optimal and suboptimal ALRT-based algor-
where Es(ik )f.g is nothing but a finite summation over all the
ithms applicable to identify both linearly modulated and
Mi possible constellation points of the ith modulation,
FSK signals under various conditions will be presented
divided by Mi , for the kth interval. Furthermore
[Note 1], in addition to ALRT-based classifiers specific to
linear modulation and FSK, respectively. Suboptimal classi- ð kT
fiers are obtained based on the approximations of the LFs at RðiÞ
k ¼ rðtÞsðiÞ
k ðtÞuT ðt ðk 1ÞT Þ dt
low SNR. Interestingly, several FB classifiers are shown to ðk1ÞT
ms (i ),2,0 1 0 0 0 0 0 0
ms (i ),2,1 1 1 1 1 1 1 1
ms (i ),4,0 1 1 0 0 0.5158 20.68 20.619
ms (i ),4,1 1 0 0 0 0 0 0
ms (i ),4,2 1 1 1 1 20.5816 1.32 1.38
ms (i ),6,0 1 0 0 0 0 0 0
ms (i ),6,1 1 1 0 0 21.5243 21.32 21.298
ms (i ),6,2 1 1 0 0 0 0 0
ms (i ),6,3 1 1 1 1 1.4897 1.96 2.22
ms (i ),8,0 1 1 1 0 25.6304 2.2 1.91
ms (i ),8,1 1 0 0 0 0 0 0
ms (i ),8,2 1 1 0 0 9.4585 22.48 22.75
ms (i ),8,3 1 0 0 0 0 0 0
ms (i ),8,4 1 1 1 1 28.5370 3.12 3.96
signal moments.
Note 6: For the definition of the nth-order, q-conjugate moment, mr,n,q(tn21) and ð21Þ
the cumulent, cr,n,q(tn21) of a stationary random process, as well the relations
between the moments and cumulants, see, e.g. [98] and [99] and Chap. 2. A 0
tn210n21 delay vector is an (n 2 1)1 vector, with all the elements equals to where p (M-PSK)(vM) and p (M -PSK)(vM) are the PDFs of the
zero. metric vM under the hypothesis that M-PSK and M 0 -PSK
IET Commun., Vol. 1, No. 2, April 2007 143
PSK modulated
signal
BPSK Yes No
ν 4, PSK > η(νQPSK
4
, M ' PSK )
M'>4
QPSK
PSK , M ' PSK )
Yes ν8, PSK > η(8
ν8 No
M '>8
8-PSK 16-PSK,
32-PSK, etc.
Fig. 2 Binary decision tree for PSK signals using the quasi-ALRT classifier
are the modulation formats of the incoming signal, respect- By comparing (20) and (23) with (13), one can say that
ively. A closed-form solution was derived in [5] for classify- the complexity of a quasi-ALRT classifier is much less
ing PSK signals with unknown carrier phase. Under the than that of the ALRT classifier, as it needs neither an
assumption of a large number of available symbols K, this averaging operation nor the computation of the Bessel
threshold was approximated by [Note 7] function.
0
-PSKÞ Quasi-ALRT with unknown carrier phase and timing offset:
hðM-PSK;M ¼ KS M=2 T M =2 ð22Þ
vM In AWGN, with vi ¼ [u 1 fs(i) K †
k gk¼1] , where the carrier phase
u and timing offset 1 are uniformly distributed over [2 p, p)
As one can notice, the threshold depends on the signal power, and [0, 1), respectively, the following statistic was used to
S. In the sequel we denote it by hQA,TE , where QA and TE distinguish between M-PSK and M 0 -PSK [5]
stand for quasi-ALRT and theoretical, respectively. An
example of a binary decision tree classifier used for PSK D1 X
X K
M
signals is shown in Fig. 2. nM;PSK ¼ rk ðd=DÞ ð24Þ
As previously explained, the nth order/q-conjugate d¼0 k¼1
moments for QAM signals do not have the attractive prop- Ð
erty used to devise a binary decision tree for PSK signal with rk(1d) ¼ kTþ1 dT
(k21)T þ 1dT r(t)uT(t 2 (k 2 1)T 2 1d T ) dt.
identification. Using (19) and the results given in Table 2, Similar to Fig. 2, a binary decision tree classifier was
it can be easily noticed that the lowest order statistics employed for PSK signal classification, with (24) compared
which can be used for QAM signals classification is of against a threshold. The threshold was empirically set,
order n ¼ 4 (q ¼ 0, 2). Following the same procedure as following the histogram method.
in the example given in the Appendix for BPSK and ALRT with unknown signal level: An ALRT algorithm was
QPSK signals, it can be easily shown that the lowest developed in [8] to identify PSK signals in AWGN, with
order metric which can be used to distinguish between vi ¼ [a fs(i) K †
k gk¼1] , where the signal level a is a
any two QAM signals is given by Rayleigh-distributed r.v. The decision rule given in (8)
was employed, with the threshold hA set to one.
X K X K
4 4
n4;QAM ¼ A rk þ B jrk j ð23Þ Miscellaneous algorithms: Numerical calculation of the
k¼1 k¼1
integrals in ALRT using a Markov chain Monte Carlo
method was performed in [24]. In the algorithms previously
where the coefficients A and B depend on the theoreti- described, AMC was treated as a hypothesis testing problem
cal values of the fourth-order/zero- and two-conjugate with a fixed number of received symbols. By formulating
moments of the QAM signals, respectively. Such a metric the AMC problem as a variable sample size hypothesis
was used in [7] to discriminate between 16-QAM and testing problem, an algorithm based on the sequential prob-
V.29, with A ¼ 0.0135 and B ¼ 20.0246. The decision ability ratio test was proposed in [25, 26].
was made by comparing the metric against a threshold,
which was empirically set. This threshold is set to maximise
the average probability of correct classification over a large 3.1.3 ALRT-AMC for FSK signals
number of data and noise realisations. It is assumed that ALRT-based classifier under the assumption of per-symbol
such simulations can be run off-line and the threshold can phase-incoherence: In AWGN, under the assumption of
be stored as a function of the noise and signal parameters. per-symbol phase-incoherence, that is vi ¼ [ffkgKk¼1
In a practical implementation, the threshold is therefore fs(i) K † K
k gk¼1] , with ffkgk¼1 as i.i.d. uniform r.v.s, the LF of
(i)
obtained from a look-up table. We denote this threshold (14), with jsk j ¼ 1, was derived in [9]. As one can easily
by hQA,E , where QA and E stand for quasi-ALRT and notice from (11) and (14), the implementation of such
empirical, respectively. classifier requires the explicit calculation of the Fourier spec-
trum of the received waveform at a set of Mi candidate fre-
p quencies [9]. The signal bandwidth of an Mi-FSK signal is
Note 7: The original presentation in [4] used 2rk/ (N0T) instead of rk in the
metric inp(20). To account for this difference, the threshold we give in (22) in defined as BWi ¼ MiTi21, with Ti as the symbol period
indeed ( (N0T )/2)M times the threshold of [4]. under the hypothesis Hi . Optimal structures were developed
144 IET Commun., Vol. 1, No. 2, April 2007
for the three possible cases: the same signal bandwidth and respectively given by [14]
distinct symbol period (equal MiTi21, with different Ti), the ( )
same symbol period and distinct bandwidth (different XK
pffiffiffi ðiÞ 2
ðiÞ ðiÞ ju 1
MiTi21, with equal Ti), and both the symbol period and LG ½rðtÞ ¼ max max Refsk rk e g 2 S T jsk j
u skðiÞ
signal bandwidth different (different MiTi21, with different k¼1
Ti) [9]. The decision was made based on the rule given in ð26Þ
(8), with hA ¼ 1.
This results in a better but more complex classifier. The The decision rule given in (8) was used, with the threshold
decision was made by comparing the metric against an set to one. An HLRT-based multi-antenna classifier was
empirical threshold, hHOC,E , where HOC and E denotes developed for BPSK/QPSK in AWGN [17, 18], with
higher-order correlation and empirical, respectively. vi ¼ [q fs(i) K †
k gk¼1] , where q is an unknown phase shift
An FSK signal classifier based also on an approximation between two adjacent antenna elements, which appears
of the LF was proposed in [13] for flat Rayleigh fading due to their spatial separation. The decision rule given in
channels. Usually, in the literature only one incoming (8) was employed, with the threshold set to one.
signal is assumed at the receiver. The case of multiple An HLRT-based classifier was developed for linear
FSK signals at the receiver was discussed in [13]. Using modulations identification in flat block fading channels in
(17), the previously discussed ALRT-based and [21], with vi ¼ [a w fs(i) K †
k gk¼1] . Similar to (27), the LF was
quasi-ALRT synchronous algorithms were transformed computed by averaging over the data symbols and using
into asynchronous classifiers, where the timing offset 1 the ML estimates of the channel phase w and amplitude
was assumed uniformly distributed over [0, 1) [10, 12, 13]. a. In [21] the threshold hH used for decision was set to
one. HLRT has the advantage over ALRT that no prior
PDFs of the channel parameters are needed, and therefore,
3.2 GLRT- and HLRT-based algorithms it is applicable to different environments, e.g. Rician and
Rayleigh fading [21].
As the ALRT algorithm suffers from high computational
complexity in most practical cases, GLRT and HLRT algor- Quasi HLRT classifier: HLRT does not seem to be a good
ithms have been investigated as possible solutions to ident- solution with an increased number of unknown parameters,
ify linear modulations [14 – 22]. In AWGN and with as finding their ML estimates can be very time consuming.
vi ¼ [u fs(i) K †
k gk¼1] , the LF for GLRT and HLRT are Quasi-HLRT classifiers, which use low-complexity yet
IET Commun., Vol. 1, No. 2, April 2007 145
accurate estimates, can be used instead. Such classifier were 4.1 FB algorithms to distinguish between different
proposed in [22] for QAM signal identification in block classes
fading channels. The estimators used for the channel ampli-
tude and phase are given, respectively, by 4.1.1 Instantaneous amplitude, phase and frequency-
based algorithms: The most intuitive way to identify the
^ 2r;2;1 ð0Þ m
a^ 4i T 4 ¼ ½2m ^ r;4;2 ð03 Þð2m2sðiÞ ;2;1 msðiÞ ;4;2 Þ1 ð28Þ modulation class of the incoming signal is to use the infor-
mation contained in its instantaneous amplitude, phase and
frequency. To extract such information, different methods
and were applied in the literature [27– 40]. The following differ-
! ences between signal classes were employed for classifi-
X
K
cation in [27– 31]:
w^ ¼ 4 1
angle rk4 ð29Þ
k¼1 † FSK signals are characterised by constant instantaneous
amplitude, whereas ASK signals have amplitude fluctu-
Then, using these estimators, the LF was calculated accord- ations, and PSK signals have information in the phase.
ing to [Note 2] The maximum of the discrete Fourier transform (DFT) of
centred (the term ‘centred’ specifies that the average is
Y
K n h removed from the data set) normalised instantaneous ampli-
LðiÞ
quasi-H ½rðtÞ ¼ EsðiÞ exp 2a^ i N01 RefsðiÞ
k rk e
jw^
g tude was used as a feature to distinguish between FSK and
k
k¼1 ASK/ PSK classes
io † ASK and BPSK signals have no information in the absol-
a^ 2i TN01 jsðiÞ
k j
2
ð30Þ ute phase, whereas M-PSK (M . 2) has. The variance of
absolute centred normalised phase was used to distinguish
The decision rule is given in (8), with the threshold set to between M-PSK (M . 2) and real-valued constellation,
one. A multi-antenna quasi-HLRT classifier was also pro- BPSK and ASK
posed in [22], with (28) and (29) used to estimate the † ASK signals have no phase information by their nature,
channel phase and amplitude on each branch, respectively. whereas BPSK has. Variance of direct (not absolute)
The threshold used for decision was set to one. centred normalised phase was used to distinguish between
BPSK and ASK classes. A binary decision tree structure
was employed to discriminate between classes, and further-
4 Feature-based approach to AMC more, within each class, as we will briefly mention later. At
each node of the tree, the decision was made by comparing a
The design of a FB algorithm first needs some features for statistic against a threshold.
data representation and then decision making [100].
Examples of features are the variance of the centred normal- In [32] and [33], the variance of the zero-crossing interval
ised signal amplitude, phase and frequency [27], the var- was used as a feature to distinguish FSK from PSK and the
iance of the zero-crossing interval [32, 33], the variance unmodulated waveform (UW). The zero-crossing interval is
of the magnitude of the signal wavelet transform (WT) a measure of the instantaneous frequency, and it is a stair-
after peak removal [36– 38], the phase PDF [44– 46] and case function for FSK signals, whereas a constant for UW
its statistical moments [47 – 49], moments, cumulants, and and PSK signals. The AMC is treated as a two hypothesis
cyclic cumulants of the signal itself [41 – 43, 53, 54, 58 – testing problem: H1 for FSK and H2 for UW and PSK.
66] etc. The entropy [67, 68], fuzzy logic [69, 70], a The hypotheses are formulated based on the Gaussian
moment matrix technique [71, 72] and a constellation assumption for the estimated feature, that is N(mHi , sH2i),
shape recovery method [73] were also used for AMC. i ¼ 1, 2, with the hypothesis-dependent mean mHi and var-
Different methods were employed for decision making, iance sH2i (the mean is actually the theoretical value of the
such as PDF-based [41 – 53], the Hellinger distance [74, feature under Hi whereas the variance is estimated under
75], the Euclidian distance [60 – 65] and unsupervised clus- each hypothesis.). An LRT is used for decision, which
tering techniques [76, 77]. Table 3 summarises most of the due to the Gaussian assumption is simplified to
FB-AMC work, emphasising the selected features, type of the comparison of the feature with a threshold h, derived
modulations, channel and the unknown parameters. In its from the LRT. For any two class problem, assuming equal
first part, algorithms which employ information extracted priors, the average probability of error is then given by
from the instantaneous amplitude, phase and frequency of
Pe ¼ ½erfcððh mH1 Þ=sH1 Þ þ erfcððmH2 hÞ=sH2 Þ=2 ð31Þ
the received signal are presented. The second and third
parts include classifiers based on the wavelet transform
and signal statistics, respectively. Finally, a classifier where erfc(.) is the complementary
Ð error function defined as
based on spectral properties of FSK signals is mentioned. erfc(x) ¼ (2p)21/2 1 2
x exp(2u /2) du.
In what follows, these FB-AMC algorithms are presented The variance of the instantaneous frequency was also
[Note 1] with a hierarchical approach in mind, that is the employed in [34, 35] to discriminate FSK from UW and
modulation class of the incoming signal is first identified PSK. In fact, the autoregressive spectrum modelling was
(e.g. ASK, PSK, QAM, FSK), and then the modulation used to extract the instantaneous frequency. The decision
order M within the recognised class. A single signal in was made by comparing the feature against a threshold.
AWGN, with the parameters perfectly known, and a rec-
tangular pulse shape uT(t) were assumed, except otherwise 4.1.2 Wavelet transform-based algorithms: The
mentioned. In addition to ASK, PSK, QAM and FSK, utility of the wavelet transform to localise the changes in
the identification of other modulations was examined the instantaneous frequency, amplitude and phase of the
in the literature, for example minimum shift keying received signal was also studied for AMC. The distinct
(MSK) [66, 78 – 80], offset QPSK (OQPSK) [79], behaviour of the Haar WT (HWT) magnitude for PSK,
continuous-phase FSK (CPFSK) [81]. QAM and FSK signals was employed for class
146 IET Commun., Vol. 1, No. 2, April 2007
Table 3: Summary of feature-based classifiers
Azzouz and Nandi [27, 28] maximum power spectral density of 2ASK, 4ASK, BPSK, QPSK, 2FSK, 4FSK – AWGN
normalised centred amplitude, standard
deviations of normalised centred amplitude,
phase and frequency
Soliman and Hsue [32, 33] variance of the zero-crossing interval UW, BPSK, QPSK, 8PSK, BFSK, 4FSK, – AWGN
sequence, phase difference and 8FSK
zero-crossing interval histograms
These modulations have been used in the simulations in the original papers
147
identification in [36 –38]. For a PSK signal this is a con- identification. In the high-SNR region, M-PSK exibits M
stant, with peaks occurring at phase changes. On the other distinct modes, while when the SNR decreases or M
hand, because of the frequency and amplitude variations increases, the peaks smear off and finally the PDF con-
in FSK and QAM, respectively, the HWT magnitude is a verges to a uniform PDF [48]. Specifically for PSK signal
staircase function with peaks at phase changes. These classification, an approximation using the Tickhonov PDF
peaks do not provide useful information for non-continuous and a Fourier series expansion of the phase PDF were
phase FSK signals. If only the phase is retained for a QAM employed in [44– 46], with a log-likelihood ratio test for
signal, it behaves like a PSK signal and thus, the HWT mag- decision. By using these methods to compute the phase
nitude is constant. On the other hand, as PSK and FSK PDF, closed-form expressions for the phase statistical
signals are of constant amplitude, amplitude normalisation moments were derived, and the PDFs of the sample esti-
has no effect on their HWT magnitude. After peak mates of the moments were used for decision making
removal, the variance of the HWT magnitude with ampli- [47 – 50]. The distribution of the sample estimate of the
tude normalisation was used to discriminate FSK from nth-order moment was assumed to be Gaussian,
PSK and QAM. Furthermore, the variance of the HWT N(mn,Hi ,sn 2,Hi), where the mean mn,Hi and variance s2n,Hi
magnitude without amplitude normalisation was employed depend on the hypothesis Hi and n. The decision criterion
to distinguish between QAM and PSK. The decisions was further reduced to comparing the sample estimates of
were made by comparing the features against some the phase moments with a threshold. The histogram of the
thresholds, chosen based on the statistical analysis of the phase difference between two adjacent symbols was used
features, to minimise the probability of error for PSK in [32, 33, 39] for PSK order identification, with the
signals [36 – 38]. decision made based on the comparison of the histogram
against particular patterns. The periodic components of
4.1.3 Signal statistics-based algorithms: To discrimi- the phase PDF were analysed for PSK order identification
nate among BPSK, ASK, M-PSK (M . 2) and QAM, the in [51], using the DFT of the phase histogram. In other
cumulant-based feature cr,4,0(03)/c2r,2,1(0) was proposed words, the empirical characteristic function of the phase
in [41], where cr,n,q(0n21) [Note 6] is the nth-order/ was exploited for classification in this work. Furthermore,
q-conjugate cumulant of the output of the matched filter in [52] the algorithm was extended to QAM signal classifi-
frkgKk¼1, at the zero delay vector. For decision, an LRT cation, by exploiting the additional information provided by
based on the PDF of the sample estimate of the feature the magnitude of the received signal. Other features
was formulated to achieve minimum probability of error. extracted from the instantaneous amplitude and phase
The moment-based feature mr,6,3(05)/m3r,2,1(0) was used in were investigated for PSK and QAM identification in [40,
[42], where mr,n,q(0n21) is the nth-order/q-conjugate 78, 83, 84], such as the kurtosis of the amplitude.
moment of the output of the matched filter frkgKk¼1, at the
zero delay vector. The goal was to distinguish between 4.2.2 Wavelet transform-based algorithm: Different
PSK and QAM. A joint power estimation and classification PSK signals give rise to different sets of peak values in
was performed in [42]. The decision was made based on the the magnitude of the Haar wavelet transform. The histo-
minimum absolute value of the difference between the gram of the peak magnitudes was employed to identify
sample estimate and prescribed values of the feature. the order of a PSK signal in [37], with the decision made
Reference [43] combined several normalised moments by comparing the histogram with the theoretical PDFs
and cumulants for training a neural network to identify corresponding to different orders.
FSK, PSK and QAM in multipath environments.
4.2.3 Signal statistics-based algorithms: Cumulant-
4.1.4 Miscellaneous algorithms: Neural networks based features were proposed in [41] to identify the order
(NNs) were also used for classification in [27, 29 – 31, 35]. of ASK, PSK, and QAM modulations, as follows: the
The Wigner – Ville distribution was used in [82] to dis- normalised cumulant of fourth-order/two-conjugate,
tinguish between PSK and FSK signals. cr,4,2(03)/c2r,2,1(0), for ASK, the magnitude of the normalised
cumulant of fourth-order/zero-conjugate, jcr,4,0(03)j/
4.2 FB algorithms for linearly modulated signals jc2r,2,1(0)j, for PSK (M . 2), and the normalised cumulant
of fourth-order/zero-conjugate, cr,4,0(03)/c2r,2,1(0), for
Classifiers summarised in Table 3, which can be applied to QAM. The theoretical values of the nth-order/q-conjugate
identify the modulation order M of linear modulations, are cumulant cs (i ),n,q , q ¼ 0, . . . , n/2, n even, for several linear
discussed in the sequel. modulations are given in Table 4. These values were com-
puted using the moment to cumulant formula [Note 6] in
4.2.1 Instantaneous amplitude and phase-based which the nth-order moments were calculated as ensemble
algorithms: Information extracted from the instantaneous averages over the noise-free unit-variance constellations
amplitude and phase of the received signal was exploited for with equiprobable symbols. Note that owing to the symmetry
linear modulation recognition, as follows. The variance of of the signal constellations considered, the nth-order
the absolute value of the normalised centred instantaneous moments for n odd are zero and hence, using the moment
amplitude was used to distinguish between 2-ASK and to cumulant formula, it is easy to show that the nth-order
4-ASK, as for the former the amplitude changes between cumulants for n odd are also zero. However, for n even we
two levels, equal in magnitude and opposite in sign, so, it have cs (i ),n,q ¼ cs (i ),n,n2q . An LRT was formulated based on
has no information in the absolute amplitude, whereas it the PDFs of the sample estimates of features, which are
has for the latter [27 – 31]. The statistic was compared Gaussian, that is N(mHi , sHi2). With a simplifying approxi-
against a threshold for decision making at a tree node, as mation, that is equal variances under all the hypotheses,
part of the binary decision tree classifier mentioned in the decision was further reduced to comparing the sample
Subsection 4.1.1. The phase PDF and its statistical estimate of the chosen feature v̂ against a threshold, with
moments were investigated for PSK signal recognition in v as any of the cumulant-based features previously men-
[44– 50]. The phase PDF is multimodal, and the number tioned. For an Nmod hypothesis testing problem, with the
of modes provides information for the PSK order hypotheses ordered such that mH1 , mH2 , , mHNmod ,
148 IET Commun., Vol. 1, No. 2, April 2007
Table 4: Theoretical cumulants for several unit 2 variance signal constellations with equiprobable symbols
cs (i ),2,0 1 0 0 0 0 0 0
cs (i ),2,1 1 1 1 1 1 1 1
cs (i ),4,0 22 1 0 0 0.5158 20.68 20.619
cs (i ),4,1 22 0 0 0 0 0 0
cs (i ),4,2 22 21 21 21 1.4148 20.68 20.619
cs (i ),6,0 16 0 0 0 0 0 0
cs (i ),6,1 16 24 0 0 1.0683 2.08 1.7972
cs (i ),6,2 16 0 0 0 0 0 0
cs (i ),6,3 16 4 4 4 2.2551 2.08 1.7972
cs (i ),8,0 2272 234 1 0 3.7797 213.9808 211.5022
cs (i ),8,1 2272 0 0 0 0 0 0
cs (i ),8,2 2272 34 0 0 2.1993 213.9808 211.5022
cs (i ),8,3 2272 0 0 0 0 0 0
cs (i ),8,4 2272 234 233 233 3.7797 213.9808 211.5022
the decision rule is to choose Hi if received signal, respectively, were used in [55] to distinguish
between BPSK and QPSK. In [56, 57] the same property was
ðmHi1 þ mHi Þ=2 , v^ , ðmHi þ mHiþ1 Þ=2 ð32Þ explored for a baseband signal. By increasing the order of the
nonlinear signal transformation beyond fourth powers, this
where mH0 ¼ 21 and mHNmodþ1 ¼ 1. argument can be extended to identify modulations of order
Note that the cumulant-based features cr,4,2(03)/c2r,2,1(0) higher than QPSK. Note that the quasi-optimal algorithm
and jcr,4,0(03)j/jc2r,2,1(0)j do not depend on a fixed carrier derived within the LB framework for PSK signal classifi-
phase u, as for q ¼ n/2 the exponential factors which cation also exploits such a property, by using the information
depend on u cancel each other, whereas for q = n/2 the extracted in time domain [3–5]. However, the signal cyclo-
phase dependency is dropped by taking the magnitude. stationarity is not exploited in this work, as the sampling is
This work was extended in [53] to classify linear modu- performed at the symbol rate T 21.
lations in frequency-selective channels. The blind alphabet- Cyclic-cumulant (CC) based features of different orders
matched equalisation algorithm [101], which was used for were investigated for modulation classification in [58–65].
equalisation, was also employed for classification. Some A feature based on fourth-order/two-conjugate and
other cumulant-based features were added [30] to the set second-order/one-conjugate CCs at the CF equal to the
of features extracted from the instantaneous amplitude, symbol rate, similar to the one that used moments [54],
phase and frequency [27– 29], to include QAM signals in was proposed in [58] and [59], to identify the order of
the set of candidate modulations to be recognised. QAM modulations. The same decision criterion as in [54]
Signal moments were applied to distinguish between was employed. In [61] a generic algorithm was proposed to
QPSK and 16-QAM in [54]. Specifically, a linear exploit signal cyclostationarity for classification. A feature
combination of the fourth-order/two-conjugate moment vector was proposed, whose components were the magni-
and the squared second-order/one-conjugate moment tudes of the CCs up to the nth-order, raised to the power of
were employed, with the coefficients and the delay vector 2/n, when n goes to infinity, and computed at all possible
optimised to maximise the probability of correct classifi- CFs and delay vectors. Apparently, such a classifier is hard
cation. A set of features was chosen for certain values of to implement. Note that raising the nth-order CC magnitudes
the delay vector, and classification was made based on the to the power of 2/n forces the features to take values within
correlation between the sample estimate and theoretical the same order of magnitude. Therefore, the classical
feature vectors. The signal-moment feature mr,6,3(05)/ Euclidian distance can be used for decision.
m3r,2,1(0) was employed to identify the order of QAM For linear modulations, the nth order/q-conjugate CC of
signals in [42], with the decision made based on the r(t), where r(t) is given in (1), with fk ¼ 0, k ¼ 1, . . . , K,
minimum absolute value of the difference between the g(t) as a raised cosine pulse shape, and n(t) as the
sample estimate and prescribed values of the feature. AWGN, and the set of CFs are given by [60, 63]
Signal cyclostationarity was also exploited for linear cðiÞ n 1 j2pb1T jðn2qÞu
r;n;q ðg; tn1 Þ ¼ a csðiÞ ;n;q T e e
modulation identification [55–64], via two approaches: spec-
tral line generation when passing the signal through different Pn1 ð1 Y
n1
nonlinearities [55–57], and periodic fluctuations with time of e j2pDf u¼1 ðÞu tu g ðÞu ðtÞ
cumulants up to the nth-order [58–65]. We note that the 1 u¼1
nth-order cycle frequencies (CFs) are given by ðÞn j2ptb
g ðt þ tu Þe dt; q ¼ 0; . . . ; n;
(n 2 2q)Df þ m/T, with m an integer [60, 63]. The nth-order
CF formula also holds for an IF signal, where Df is replaced ð33Þ
by the IF frequency, fIF . With this property, the cyclostatio-
narity of the received signal was exploited for AMC kn;q ¼ fg : g ¼ b þ ðn 2qÞDf ; b ¼ k=T ;
through a pattern of sine-wave frequencies in signal poly-
nomial transformations. For example, the 2fIF and 4fIF sinu- k integer; cðiÞ
r;n;q ðg; tn1 Þ = 0g ð34Þ
soids that appear in the second and fourth powers of the
IET Commun., Vol. 1, No. 2, April 2007 149
where g is a CF, ( )u represents a possible conjugation of the a smaller number of symbols is needed to attain a specific
uth term, u ¼ 1, . . . , n, such that the total number of performance. In [63 – 65] the minimum Euclidian distance
conjugations is q, and (2)u is the minus sign associated between the sample estimate and prescribed feature-vectors
with the possible conjugation ( )u , u ¼ 1, . . . , n 2 1. was also used for decision making. A raised-cosine pulse
Since n(t) is a stationary, zero-mean Gaussian process, its shape was considered in [60– 65].
cumulants are time independent and non-zero only for the 4.2.4 Miscellaneous algorithms: Finally, the Radon
second order. Therefore, AWGN does not have any contri- transform was investigated for QAM classification in
bution to the higher-order (n 3) CCs of r(t). One can [85 – 87].
easily notice that by taking the magnitude of the nth order
CC, a feature robust to the carrier phase and timing offset 4.3 FB algorithms for FSK signals
is obtained.
In [60 – 61, 62], the magnitudes of the CCs up to the Similar to using the information contained in the instan-
fourth- and sixth-order, at a CF equals to taneous phase to identify the order of the PSK modulation,
g ¼ (n 2 2q)Df þ 1/T and a delay vector for which a the information extracted from the instantaneous frequency
maximum is reached (i.e. tn21 ¼ 0n21 [63]), were investi- is exploited to recognise the order of the FSK modulation.
gated as features, respectively. Based on these features, a In [27– 31], the variance of the absolute value of the nor-
feature vector was proposed in [62], as malised centred instantaneous frequency was used to dis-
tinguish between 2-FSK and 4-FSK. The feature was
F ðiÞ ¼ ½jcðiÞ ðiÞ ðiÞ
r;2;0 ðg; 0Þj jcr;2;2 ðg; 0Þkcr;4;0 ðg; 03 Þj compared against a threshold for decision, at a tree node,
as part of the binary decision tree classifier mentioned in
j cðiÞ ðiÞ ðiÞ y
r;4;4 ðg; 03 Þkcr;6;0 ðg; 05 Þj jcr;6;6 ðg; 05 Þj ; Subsection 4.1.1. In [34], the instantaneous frequency
i ¼ 1; . . . ; Nmod ð35Þ derivative was used to distinguish between 2-FSK and
4-FSK, under the assumption of the same bandwidths of
The CC-based features are estimated from Kr samples, the signals. The height of the peaks which occur in the dif-
taken over the observed K symbol interval [102]. Note ferentiated instantaneous frequency is proportional to the
that the received signal is oversampled in order to exploit frequency deviation, and thus, for 4-FSK this is expected
signal cyclostationarity. The sampling frequency is equal to be two times lower than for 2-FSK. If the peak average
to r/T, with r a positive integer, called the oversampl- falls below a certain threshold, 4-FSK is chosen, otherwise
ing factor. The decision is made by comparing the 2-FSK. As for PSK order identification, the number of
sample estimate with prescribed feature vectors from a modes in the instantaneous frequency histogram was
look-up table employed to determine the order of the FSK modulation
in [32, 33]. The number of modes in the histogram of the
^
^ı ¼ arg min dðF ðiÞ ; FÞ ð36Þ Haar wavelet transform magnitude was investigated for
1iNmod the FSK order identification in [36, 37]. If M/2þ1 to M
modes appear in the histogram, the input is identified as
where ^ı represents the decision on the modulation type of M-FSK. Finally, spectral properties of FSK signals were
the intercepted signal, F(i) is the theoretical feature-vector explored for classification in [88].
which corresponds to the ith modulation in the look-up
table, F^ is the estimated feature vector, and d(., .) is the 5 Numerical results
Euclidean distance. Joint detection and classification was
investigated in [62], with single and multiple incoming After scanning the literature to answer the question ‘which
signals present at the receiver, respectively. Multiple AMC technique is the best in terms of performance under
signals, which overlap in time and frequency but have dis- realistic conditions?’ it turned out that performance compari-
tinct symbol rates and hence, different cycle frequencies, son of published classifiers is not straightforward. There are a
can be distinguished using CCs (selectivity property of number of reasons for this. First, performance of different
CCs) [103]. Moreover, it is claimed that all the parameters classifiers cannot be compared, unless the candidate modu-
necessary for identification, i.e. symbol period, carrier lations are the same. Second, most of the classifiers are
frequency offset, excess bandwidth and signal amplitude designed to handle specific unknown parameters. So, one
are estimated. However, no details are given about the cannot really compare their performance, unless the uncer-
estimation methods. Eight-order CC-based features, that is tainties the classifiers take into account are the same.
the magnitude of the eighth-order (q ¼ 0, . . . , 8) CCs at Nevertheless, to obtain some insight, we make some com-
the CF g ¼ 1/T þ (n 2 2q)Df and zero delay vector were parisons in Subsections 5.1–5.3. We rely on some numerical
investigated in [63] for classifying real- and complex- results reported in the literature. However, we have also
valued constellations, respectively. The nth-order (n ¼ 4, simulated some of the algorithms under the same conditions,
6, 8) CC-based features were shown to be robust to a in Subsections 5.1–5.3 to make the comparison possible.
carrier frequency offset and phase jitter for q ¼ n/2 and
used for QAM classification in [64]. Features similar to 5.1. Comparative study of classifiers for PSK
those proposed in [63] were investigated in [65] for classify- signals
ing linear modulations in block fading channels. In this
case, the features were extracted from the signal at the Performance achieved with several algorithms for classify-
output of a selection combiner. Such a CC-based multi- ing PSK signals in AWGN is presented in Table 5. We
antenna classifier takes advantage of the robustness of the consider here the ideal scenario, that is no unknown par-
CC-based features to phase, timing errors and stationary ameters, as well as the scenarios with unknown carrier
noise, as well as of the ability of the selection combiner phase, and unknown carrier phase/timing offset, respect-
to mitigate the impact of fading via spatial diversity. ively. We have examined the ALRT, quasi-ALRT, HLRT,
Furthermore, the classifier is robust to the variations of and cumulant-based algorithms. For illustration, BPSK
the Ricean factor and a possible correlation among the and QPSK are considered as candidate modulations. Of
antennas. By increasing the number of antenna elements, course, when higher order modulations are included in the
150 IET Commun., Vol. 1, No. 2, April 2007
Table 5: Performance and details of several classifiers for BPSK against QPSK
These are not included when designing the classifier, and their impacts on the performance were studied to evaluate the classifier
robustness.
These numerical results are not given in the papers where the classifiers are proposed, and we have simulated them to perform a
comparison with other methods.
†
The maximisation procedure with respect to the carrier phase u was carried out by discrete search with a 1 degree resolution. Owing to
the symmetry of the investigated constellations, the maximisation procedure can be performed over [0, p/2], instead of [0, 2p] [14, 21, 22].
modulation pool, higher SNRs and/or a larger number of As expected, the ALRT-based classifier designed for an
symbols are needed to achieve the same performance ideal scenario fails under an unknown carrier phase offset.
[2, 5, 21]. We have simulated these modulations to draw Investigated solutions are the HLRT, quasi-ALRT and
some basic yet insightful conclusions, which shed some cumulant-based classifiers. A slightly better performance
light on major techniques. is achieved with HLRT, when compared with the
Subsequently, the probability of correct classification quasi-ALRT. Also, the quasi-ALRT provides a better
Pcc is used to evaluate the classification performance, with performance than the cumulant-based classifier. However,
P(iji)
c estimated based on 1000 Monte Carlo trials. Unless a synchronisation error dramatically degrades the
otherwise mentioned, the pulse shape is rectangular, the performance of the former, whereas it has a smaller effect
received SNR per symbol is defined as gs ¼ ST/N0 , the on the latter. Though, the quasi-ALRT can be modified as
symbol period and signal power are set to T ¼ 1 and in (24) to further account for the unknown timing.
S ¼ 1, respectively and the number of symbols is K ¼ 100. Performance enhancement can be achieved by using
Under ideal conditions, the ALRT-based classifier in (10) multiple receive antennas, even with L ¼ 2. More examples
and (8), with hA ¼ 1, provides a Pcc of 0.975 at 23 dB SNR of modulation recognition with multiple antennas can be
(5-1) [Note 8]. A 3 dB SNR improvement is achieved with found in [22, 65].
an extra antenna, i.e. L ¼ 2, and maximal ratio combining at
the receiver, presented in (12) [Note 3] and (8), with
5.2 Comparative study of classifiers for QAM
hA ¼ 1(5-2). With a carrier phase u uniformly distributed
signals
over [2p, p), the performance of the ALRT-based classifier
designed for no unknown parameters, shown in (10) and (8) The performance of several algorithms is given in Table 6,
with hA ¼ 1, drops to 0.63 at 23 dB (5-3). With the when discriminating between 16-QAM and V.29 signals in
quasi-ALRT classifier in (20) and (22), with M ¼ 2, a Pcc AWGN. We consider here the ideal scenario, as well as
of 0.96 at 22 dB SNR is achieved (5-4), which is close to the case with the unknown carrier phase u. We have investi-
that under ideal condition (5-1), (5-4). As noticed from gated the ALRT, quasi-ALRT, HLRT, quasi-HLRT,
(5-5), a slightly better performance is achieved with the cumulant- and cyclic cumulant-based algorithms. Under
HLRT designed for an unknown carrier phase u, given in ideal conditions, ALRT in (10) and (8), with hA ¼ 1, recog-
(27) and (8), with the threshold set to 1. Using the cumulant- nises 16-QAM and V.29 with a Pcc of 0.99 at 7 dB SNR
based classifier in (32) with v̂ ¼ j^cr,4,0(03)/^c2r,2,1(0)j (for the (6-1). With an unknown carrier phase u, uniformly distribu-
cumulant estimator formula one can see [41], (5) and (6)), ted over [2p, p), the ALRT classifier designed for ideal
Nmod ¼ 2, mH1¼1 and mH2 ¼ 2, 4 dB SNR is required to conditions fails (6-2), as expected. For a detailed sensitivity
attain a Pcc of 0.96 (5-6). However, a timing error degrades analysis of the ALRT-based classifier to model mismatches
the performance of the quasi-ALRT classifiers dramatically, one can see [93]. A 2 dB SNR degradation occurs
that is a 13 dB SNR loss (5-4), (5-7), whereas only 2 dB loss when using the cumulant-based algorithm in (32), with
is observed for the cumulant-based classifier (5-6), (5-8). v̂ ¼ c^ r,4,0(03)/^c2r,2,1(0), Nmod ¼ 2, mH1 ¼ 20.68, and mH2 ¼
0.5185, under ideal conditions (6-3). This also fails when
Note 8: When necessary, we refer to the ith row in Table 5, Table 6, Table 7 as the unknown carrier phase u appears (6-4). Alternatively,
(5-i), (6-i), and (7-i), respectively. even the quasi-ALRT which is designed for an unknown
IET Commun., Vol. 1, No. 2, April 2007 151
Table 6: Performance and details of several classifiers for 16QAM against V.29
These are not included when designing the classifier, and their impacts on the performance were studied to evaluate the classifier
robustness.
These numerical results are not given in the papers where the classifiers are proposed, and we have simulated them to perform a
comparison with other methods.
†
The maximisation procedure w.r.t. the carrier phase u was carried out by discrete search with a 1 degree resolution. Owing to the
symmetry of the investigated constellations, the maximisation procedure can be performed over [0, p/2], instead of [0, 2p] [14, 21, 22].
þ
The values of these thresholds were not specified in the papers.
þþ
The LF in (30) simplified for the AWGN channel, considered in this simulation, with the carrier phase as the only unknown parameter,
and ai replaced by S, which is assumed perfectly known.
carrier phase scenario (see (23), with A ¼ 0.0135, with enough symbols, the CC-based classifier provides
B ¼ 20.0246, and the threshold hQA,E [Note 9], provides a superior performance when compared to the quasi-HLRT,
Pcc of 0.88 at 30 dB SNR (6-5). An acceptable performance taking also advantage of the CC properties.
is attained with the HLRT classifier in (31) and (8), with hH,E
[Note 9], designed for unknown carrier phase: a Pcc of 0.99 at
5.3 Comparative study of classifiers
9 dB SNR (6-6). By using the low complexity quasi-HLRT
for FSK signals
classifier, with the LF given in (30) [Note 10], the carrier
phase as in (29), and threshold set to 1, 19 dB SNR is Here we mainly take the numerical results from [9].
requested to attain the same performance (6-7). By increas- Performance of the optimal ALRT and quasi-ALRT algor-
ing the number of symbols to 6000, 11 dB SNR is needed ithms is given in Table 7, to distinguish between 32-FSK
to obtain a Pcc of 0.91 (6-7). Alternatively, with 6000 and 64-FSK in AWGN, under the assumption of per-symbol
symbols, the CC-based classifier in (35) and (36) provides phase incoherence and identical bandwidths of the incoming
a Pcc of 0.99 at only 9 dB (6-8). Moreover, this classifier is signals. The symbol period of 64-FSK is two times larger
robust to the timing offset, is applicable to a larger pool of than that of 32-FSK, under the assumption of the equal band-
modulations, including QAM, PSK and ASK, and benefit widths. With the same observation interval, K ¼ 10 and 5
from the selectivity property of CCs. Note that with the symbols are considered for 32-FSK and 64-FSK, respect-
CC-based classifier, the pulse shape is raised cosine, with a ively, to have a fair comparison. The optimal ALRT in
roll-off factor of 0.35 and the SNR is defined at the output (14) and (8), with hA ¼ 1, provides a Pcc of 0.975 at 6 dB
of the root raised cosine receive filter. With multiple SNR (5-1). The quasi-ALRT classifier based on the first-
receive antennas one can enhance the performance, as order correlation, that is (25) with the first three terms and
shown for quasi-HLRT [22] and CC-based classifiers [65]. the threshold hHOC,E [Note 9] for decision making, provides
We can conclude that the ALRT- and cumulant-based a Pcc of 0.74 at the same SNR (5-2). The quasi-ALRT
classifiers, designed for an ideal scenario, fail when the classifier that employs the first- and second-order
unknown carrier phase offset is present. Investigated sol- correlations, that is the first five terms in (25), provides a
utions are the quasi-ALRT, HLRT, quasi-HLRT, and Pcc of 0.90 (5-3). The ALRT-based classifier essentially
CC-based classifiers. With the quasi-ALRT classifier, breaks down for a frequency drift of half of the frequency
QAM signals are not identified accurately enough. As deviation, that is Df ¼ fd/2 (5-4), whereas the quasi-ALRT
expected, the HLRT classifier provides a better classifi- classifier is insensitive to such a model mismatch [9].
cation performance compared with the quasi-HLRT. By The performance of the quasi-ALRT classifier
increasing the number of processed symbols, the perform- approaches to the optimal ALRT, by increasing the
ance of the quasi-HLRT classifier improves. However,
Note 10: The LF in (30) is simplified for the AWGN channel, considered in this
simulation, with the carrier phase as the only unknown parameter, and âi
Note 9: The values of these thresholds were not specified in the papers. replaced by S, which is assumed perfectly known.
threshold
3 quasi-ALRT (second-order correlation), phase jitter fwkgKk¼1 – 6 0.90 10
(25) with the first five terms,
expressed for K(K/2) symbols for
32FSK (64FSK) and hþ
HOC,E as the
threshold
4 ALRT (see the first row of this table) phase jitter fwkgKk¼1 Frequency offset (Df ¼ fd/2) 6 0.57 10
These are not included when designing the classifier, and their impacts on the performance were studied to evaluate the classifier
robustness. þThe values of these thresholds were not specified in the papers
correlation order. The quasi-ALRT classifier has the advan- development of classification methods which rely less on
tage of robustness to the carrier frequency offset. preprocessing is another topic for further investigation.
New classification problems have raised as a result of emer-
ging wireless technologies, such as, single carrier against
6 Conclusion multicarrier modulation recognition, classification of
signals received from single and multiple transmit antennas,
Based on a comprehensive literature survey, this paper has identification of space-time modulation format, etc. These
summarised the two main approaches to automatic modu- issues mean that AMC in real-world environments con-
lation classification (AMC), that is the likelihood-based tinues to be a dynamic research field.
(LB) and the feature-based (FB) methods, and has lightened
their advantages and drawbacks. Although the LB approach
7 References
provides optimal performance, it is difficult to obtain an
exact analytical solution for the decision function, when 1 Sills, J.A.: ‘Maximum-likelihood modulation classification for PSK/
the number of unknown parameters increases. When a QAM’. Proc. IEEE MILCOM, 1999, pp. 57– 61
closed-form solution exists, the computational complexity 2 Wei, W., and Mendel, J.M.: ‘Maximum-likelihood classification for
can make the classifier impractical. By using a low SNR digital amplitude-phase modulations’, IEEE Trans. Commun., 2000,
48, pp. 189– 193
approximation of the likelihood function (LF), the so-called 3 Kim, K., and Polydoros, A.: ‘Digital modulation classification: the
quasi-ALRT algorithms were proposed in the literature. BPSK versus QPSK case’. Proc. IEEE MILCOM, 1988,
These algorithms provide near optimal performance for pp. 431–436
identifying PSK and FSK signals. Nevertheless, they are 4 Polydoros, A., and Kim, K.: ‘On the detection and classification of
quadrature digital modulations in broad-band noise’, IEEE Trans.
not accurate for recognising QAM signals. Using ML esti- Commun., 1990, 38, pp. 1199–1211
mates of the unknown quantities, GLRT and HLRT tech- 5 Huang, C.Y., and Polydoros, A.: ‘Likelihood methods for MPSK
niques were investigated as two alternatives. Although modulation classification’, IEEE Trans. Commun., 1995, 43,
GLRT has some advantages, it fails in identifying nested pp. 1493–1504
constellations. However, HLRT does not have this 6 Sapiano, P.C., and Martin, J.D.: ‘Maximum likelihood PSK
classifier’. Proc. ICASSP, 1996, pp. 1010–1014
problem. However, with several unknown parameters, 7 Long, C., Chugg, K., and Polydoros, A.: ‘Further results in likelihood
finding the ML estimates can be very time consuming. classification of QAM signals’. Proc. IEEE MILCOM, 1994,
The complexity is reduced in quasi-HLRT classifiers, pp. 57– 61
which rely on low-complexity yet accurate estimators. 8 Hong, L., and Ho, K.C.: ‘Classification of BPSK and QPSK signals
Obviously, there is a trade-off between the complexity with unknown signal level using the Bayes technique’. Proc. IEEE
ISCAS, 2003, IV.1–IV.4
and performance, which depends on the estimation 9 Beidas, B.F., and Weber, C.L.: ‘Higher-order correlation-based
method. In the FB approach, some signal features are approach to modulation classification of digitally
employed to identify the modulation format. Although sub- frequency-modulated signals’, IEEE J. Sel. Areas Commun., 1995,
optimal, FB algorithms can be simpler to implement. Many 13, pp. 89–101
10 Beidas, B.F., and Weber, C.L.: ‘Asynchronous classification of
AMC algorithms which use the instantaneous amplitude, MFSK signals using the higher order correlation domain’, IEEE
phase, frequency, signal wavelet transform, and signal stat- Trans. Commun., 1998, 46, pp. 480– 493
istics such as moments, cumulants, and cyclic cumulants, 11 Beidas, B.F., and Weber, C.L.: ‘General framework for the
were compactly presented, in addition to other miscella- high-order correlation domain’. Proc. IEEE MILCOM, 1995,
neous techniques. In a hierarchical classification system, pp. 180–185
12 Beidas, B.F., and Weber, C.L.: ‘Higher-order correlation-based
some of these features can be used to identify the modu- classification of asynchronous MFSK signal’. Proc. IEEE
lation class and then, within each class, the modulation MILCOM, 1996, pp. 1003– 1009
order can be determined. 13 El-Mahdy, A.E., and Namazi, N.M.: ‘Classification of multiple
Accurate preprocessing is required for the effective M-ary frequency-shift keying signals over a Rayleigh fading
channel’, IEEE Trans. Commun., 2002, 50, pp. 967 –974
implementation of most of the known AMC algorithms. 14 Panagiotou, P., Anastasoupoulos, A., and Polydoros, A.: ‘Likelihood
Devising low complexity blind algorithms for joint par- ratio tests for modulation classification’. Proc. IEEE MILCOM,
ameter estimation is a topic of interest in AMC. In addition, 2000, pp. 670–674
pp. 1878–1881
N01 ja S ejw sðiÞ 2
k ðtÞuT ðt ðk 1ÞT Þj dt
0 k¼1
86 Wood, S.L., and Treichler, J.R.: ‘Performance of the Radon
transform method for constellation identification’. Proc.
ASILOMAR, 1988, pp. 119–123 and after some manipulations, becomes
87 Wood, S.L., Ready, M.J., and Treichler, J.R.: ‘Modem constellation
identification: A performance comparison of two methods’. Proc. LðiÞ
A ½rðtÞ
ICASSP, 1990, pp. 1651– 1654
88 Yu, Z., Shi, Y.Q., and Su, W.: ‘M-ary frequency shift keying signal Y
K h pffiffiffi n o i
classification based on discrete Fourier transform’. Proc. IEEE ¼ EsðiÞ exp 2a S N01 Re ejw RðiÞ
k a2 STN01 jsðiÞ
k ðtÞj
2
MILCOM, 2003, pp. 1167–1172 k¼1
k
89 Cimini, L.J.: ‘Analysis and simulation of a digital mobile channel
using orthogonal frequency division multiplexing’, IEEE Trans. ð37Þ
Commun., 1985, 33, pp. 665–675
90 van Nee, R., and Prasad, R.: ‘OFDM for wireless multimedia Ð
LðiÞ and
A ½rðtÞ
( !)
PK X K
2
TN01 jskðiÞ j2 ðiÞ LðQPSKÞ
¼ EfsðiÞ gK Ea ea k¼1 I0 2aN01 Rk ð40Þ A ½rðtÞ
k k¼1 k¼1
pffiffiffi 4
pffiffiffi
’ exp K 2ð S N01 Þ2 jm
^ r;2;1 j þ 121 S N01 jm^ r;4;0 j
Using the following integral
n 2
o ð þ1 2 2
pffiffiffi 4
pffiffiffi 6
Ea eK1 a I0 ðaK2 Þ ¼ eK1 a I0 ðaK2 Þ2aV1 ea =V da þ41 S N01 jm^ r;4;2 j þ 601 S N01 jm^ r;6;1 j
0
ð þ1
pffiffiffi 6
pffiffiffi 6
1 1 1 1
¼ V1
2
2aeK3 a I0 ðaK2 Þ da þ24 S N0 ^ r;6;2 j þ 36
jm S N0 ^ r;6;3 j þ ...
jm
0
1
K31 K22 ð44Þ
¼ V1 K31 e4 ð41Þ
(40) becomes The difference between the log-LFs is therefore given by
(
1
LðiÞ pffiffiffi 1 2
A ½rðtÞ ¼ EfsðiÞ gK PK
1 þ VTN0 1 ðiÞ 2 lnLAðBPSKÞ ½rðtÞ lnLðQPSKÞ ½rðtÞ ’ K S N0 ^ r;2;0 j
jm
k¼1 jsk j
k k¼1
A
"P #)
VN02 j Kk¼1 RðiÞ j 2
pffiffiffi 4
pffiffiffi 6
k 1 1 1 1
exp P ð42Þ þ3 S N0 ^ r;4;1 j þ 360
jm S N0 ^ r;6;0 j þ
jm
1 þ VTN01 Kk¼1 jsðiÞk j
2
ð45Þ
8.2 Derivation of (20) for discriminating between
BPSK and QPSK
As one can easily notice from (45), the lowest order statistic
The nth order moments of symmetric constellations are zero which can be used to PKdiscriminate between BPSK and
for n odd. Using the values of the moments ms (i ),n,q for QPSK is Kjm^ r,2,0j ¼ j k¼1 r2k j, that is n2,PSK .