See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228892343

Algorithms of digital modulation classification and their verification

Article in WSEAS TRANSACTIONS ON COMMUNICATIONS · September 2010

CITATIONS READS

7 3,291

2 authors:

Anna Kubankova David Kubanek

Brno University of Technology Brno University of Technology
15 PUBLICATIONS 55 CITATIONS 81 PUBLICATIONS 901 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by David Kubanek on 29 May 2014.

The user has requested enhancement of the downloaded file.

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

Algorithms of digital modulation classification and their verification

ANNA KUBANKOVA and DAVID KUBANEK
Department of Telecommunications
Brno University of Technology
Purkynova 118, 612 00 Brno
CZECH REPUBLIC
[email protected], [email protected]

Abstract: - Two methods of the classification of digital modulations ASK, FSK2, FSK4, MSK, BPSK, QPSK,
8PSK and QAM-16 are presented. The first algorithm is based on module and phase spectrogram analysis. Here
it is very important to estimate the symbol length of the received signal and thus we introduce two methods for
this purpose. They are based on the wavelet transform and cepstrum analysis. The second algorithm of
modulation classification uses instantaneous values of signal (instantaneous amplitude, phase and frequency). A
method of the practical verification of classification algorithms in laboratory environment is described.

Key-Words: - Classification of modulations, examination, modulation, symbol length estimation.

1 Introduction Recently, new methods of modulation

In connection with the requirement for faster and classification and symbol length estimation were
more reliable communication, digital processing proposed. Modulation classification methods are
methods and digital communications are mainly based on spectrogram analysis [3], [4] and
used. Together with the rapid growth in the last instantaneous values of signals [5]. Methods for
decade of cellular technologies, PCS (Personal symbol length estimation are based on the wavelet
Communication Services) and WLAN (Wireless transform [6] and cepstrum analysis [7].
Local Area Network) services, a number of different In this paper, all the above recognition and
wireless communication standards were proposed detection algorithms are described shortly and a
and employed, each of them having its own unique method for their examination using real signals is
modulation type, access technique, etc. To realize designed. Also, the necessary equipment for
seamless inter-communication between these measurement is described.
different systems, a multiband, multimode smart
radio system such as software radio, is becoming the
focus of commercial and research interests. The 2 Modulation Recognition Method
automatic modulation classification technique, Based on Spectrogram Analysis
which is indispensable for the automatic choice of This method is based on module and phase
the appropriate demodulator, plays an important role spectrogram analysis. The module and phase
in such a multimode communication system [1]. spectrograms were calculated with the segment size
Automatic identification of the type of digital equal to the symbol length. The obtained
modulation has found application in many areas, spectrograms are shown in fig. 1 – 2. Module
including electronic warfare, surveillance, and threat spectrograms for BPSK and QPSK modulations are
analysis [2]. identical.
ASK (Amplitude Shift Keying), MPSK (M-ary
Phase Shift Keying), MFSK (M-ary Frequency Shift
Keying), QAM (Quadrature Amplitude Modulation)
and MSK (Minimum Shift Keying) were chosen for
the analysis because they belong to the best-known
digital modulations. These modulation types are
used in modern radio telecommunication systems
(GSM, WiFi, WiMAX, etc.).

ISSN: 1109-2742 563 Issue 9, Volume 9, September 2010

WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

a) d)
Fig. 1. Module spectrograms for each modulation type

b)

a)

c)

b)

ISSN: 1109-2742 564 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

Fig. 4. Occurrence count of maxima in separate rows

of MSK spectrogram

Similar to FSK modulation, two carrier

frequencies can be found in the module spectrogram
of the MSK modulation. It is possible to detect two
maxima in the spectrogram by observing the
occurrence of column maximum amplitudes on
c) separate frequencies. However, it is not easy to
Fig. 2. Phase spectrograms for each modulation type recognize them, because the distance between them
is just half of the bit rate R/2. Moreover, in a signal
For the estimation of spectrogram features, it is
with noise, the other maxima often occur and cause
advantageous to use their histograms. The
a wrong detection of MSK modulation.
spectrogram can be presented as a matrix of
The analysis of module spectrograms described
numbers A(i, j), where i in the range [0, M-1] and j
above can distinguish three different modulations
in the range [0, N-1] are indexes of rows or columns
with varying carrier frequency. For the remaining
of the matrix. For the analysis of module
modulation types (ASK, BPSK, QPSK, 8PSK and
spectrograms it is suitable to count the occurrences
QAM-16) it is necessary to analyze the phase
of column maxima, in particular rows. The analysis
spectrograms. The graphs of maximum occurrence
results are shown in the following figures.
counts used for module spectrogram analysis do not
provide any usable properties.
From a detailed view of phase spectrograms it is
apparent that several different values of brightness
(phases) occur in the area around the carrier
frequency. The number of these values corresponds
to the number of phase positions used in the
modulation. Thus the analysis must evaluate how
many phase values occur on the carrier frequency. If
Fig. 3. Occurrence count of maxima in separate rows the carrier frequency is not known, it can be easily
of FSK2 spectrogram found from the module spectrogram. One, two, four,
eight and twelve phase values occur on the carrier
The FSK2 and FSK4 modulation types are the frequency which correspond with the theoretical
easiest for recognition, because two or four distant expectation for the ASK, BPSK, QPSK, 8PSK and
carrier frequencies occur in the module QAM-16 modulations. The counts of differences of
spectrogram. Therefore the method of finding the phase values at the carrier frequency for different
column maximum occurrence in separate rows modulations are shown in the following three
easily detects two or four characteristic maxima for figures.
these modulations.

ISSN: 1109-2742 565 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

wavelet transform coefficients for different

modulation types are shown in fig. 8 – 11.

Fig. 5. Occurrence count of phase differences at the

carrier frequency for BPSK spectrogram

Fig. 8. Wavelet coefficients (level 1) of BPSK

signal

Fig. 6. Occurrence count of phase differences at the

carrier frequency for QPSK spectrogram

Fig. 9. Wavelet coefficients (level 1) of FSK signal

Fig. 7. Occurrence count of phase differences at the

carrier frequency for QAM-16 spectrogram

The analysis results strongly depend on the

segment size used for spectrum calculation. The
Fig.10. Wavelet coefficients (level 1) of QPSK
characteristic properties of modulation types are
signal
apparent only in spectrograms that are obtained with
the segment size equal to the symbol length.
Therefore it is necessary to find the signal element
length prior to obtaining the spectrograms. For this
purpose, methods based on wavelet transform and
cepstrum analysis were suggested.

2.1 Method of Symbol Length Detection

Based on Wavelet Transform Fig.11. Wavelet coefficients (level 1) of
Signals with any of the above modulation types QAM-16 signal
were transformed by the wavelet transform. For
phase modulation types, as the frequency is In the wavelet coefficients of the FSK signal
constant, the wavelet coefficients resemble a series there is also a visible difference between states of
of spikes and the interval between neighboring signal with a suitably selected level of
peaks is the symbol period. The wavelet transform decomposition.
can enhance the high-frequency components which Thus, the received signal is transformed by the
appear when the signal phase or amplitude is level 2 wavelet transform and by the order 2
changed. A step change in the phase or amplitude Daubechies function. Because the wavelet
will show up in several coefficient levels. The coefficients of the FSK signals have a different
character than the wavelet coefficients of phase

ISSN: 1109-2742 566 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

modulations, the Hilbert transform is executed after The periodic component in the spectrum causes a
the wavelet transform. Via the Hilbert transform, a peak in the cepstrum. The cepstrums of each
two- or four-step envelope of wavelet coefficients is modulation types are shown in fig. 13 – 17.
created for the FSK signals and the coefficients
occurring when the phases of other signals change
are emphasized. Then, the functions obtained are
differentiated. Thus, similar to phase modulated
signals, the peak series is obtained from the two- or
four-step envelope of FSK signal. The distance
between the peaks matches one or several symbol
lengths. It appears from this that the peaks are
periodical, therefore the Fourier transform can be Fig.13. Cepstrum of BPSK signal
used. Several maxima are found in the spectrum, the
distance between them is obtained and the symbol
length is calculated in agreement with:

Ns 1
T= ⋅ , (1)
T peaks f s

where Tpeaks is the period of spectrum components, Fig.14. Cepstrum of MSK signal
Ns is the total number of samples in the spectrum,
and fs is the sampling frequency. The block-diagram
of the designed algorithm is shown in fig. 12.

Calculation of
Hilbert
the wavelet Differentiation
coefficients transform

Fig.15. Cepstrum of QPSK signal

Estimation of the
Symbol length Fourier
spectrum
calculation
components period
transform

Fig.12. Block-scheme of the symbol length

detection algorithm based on wavelet transform

Fig.16. Cepstrum of FSK signal

2.2 Method of Symbol Length Detection
Based on Cepstrum Analysis
The zero points in the power spectrum of digitally
modulated signals are periodic and the real cepstrum
can be used for the detection of signal element
length. The real periodic cepstrum may be
determined from the power spectrum S(f):

[ ]
c(n ) = Re{IFFT log S ( f ) }. (2) Fig.17. Cepstrum of QAM-16 signal

The position of the first peak in the cepstrum

matches the signal element length:

ISSN: 1109-2742 567 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

T = n C1 peak ⋅ T s , (3) The instantaneous phase φ(t) is given by

where nC1 peak is the position of the first peak in the
ϕ (t ) = arg{z (t )} . (7)
cepstrum and Ts is the sampling rate.
However it is a problem to find the correct
position of the first peak in the cepstrum Finally, the instantaneous frequency f(t) is given
automatically. Therefore the cepstra of the by
modulated signals were transformed by means of
1 dϕ (t )
the Hilbert transform. The signals obtained have f (t ) = . (8)
distinct periodic peaks and their spacing is the same 2π dt
as the value of nC1 peak . This spacing can be detected
if we compute the spectrum of this signal. In this The maximum value of the spectral power
spectrum, the reciprocal value of the position of the density of the normalized-centered instantaneous
first peak is proportional to the signal element amplitude γmax of the received signal is used to
length: discriminate between frequency modulations
(FSK2, FSK4 and MSK) and amplitude and phase
Ns 1 modulations (ASK, MPSK and QAM-16). [8]
T = ⋅ , (4)
n S1 peak f s
γ max = max DFT(acn (i )) / N s ,
2
(9)
where nS1peak is the position of the first peak in the
spectrum, Ns is the total number of samples in the
where Ns is the number of samples per signal and
spectrum, and fs is the sampling frequency. The
acn(i) is the value of the normalized-centered
block-diagram of the algorithm for symbol length
detection based on cepstrum analysis is shown in i
instantaneous amplitude at time instants t = , (i
fig. 18. fs
= 1, 2, …, Ns), and it is defined by
Cepstrum Fourier
a(i )
Hilbert transform
calculation transform
a cn (i ) = a n (i ) − 1 , where an (i ) = , (10)
ma

Detection of
Calculation of
spectrum line
symbol length
period where ma is the average value of the instantaneous
amplitude
Fig. 18. Block-diagram of algorithm for symbol
length detection 1 s
N
ma = ∑ a(i ) .
N s i =1
(11)

3 Modulation Recognition Method

Based on Analysis of Instantaneous Normalizing of the instantaneous amplitude is
necessary in order to compensate the channel gain.
Values of Signal
The dependence of γmax on SNR for each
The received real signal x(t) can be represented as
modulated signals is shown in fig. 19.
the analytic signal z(t), which can be expressed as

z (t ) = x(t ) + jy (t ) , (5)

where y(t) is the Hilbert transform of x(t), and j is

the imaginary unit.
The instantaneous amplitude a(t) is defined as

a (t ) = z (t ) = x 2 (t ) + y 2 (t ) . (6)

ISSN: 1109-2742 568 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

Fig. 19. Dependence of γmax on SNR Fig. 20. Power spectrum of MSK

The FSK2, FSK4 and MSK modulations have no

amplitude changes and their γmax is less than a
threshold value tr(γmax). The band-limited MPSK,
ASK and QAM-16 modulations have amplitude
changes and their γmax is larger than tr(γmax). This
threshold value has to be determined appropriately
in order to discriminate between the two modulation
groups. From fig. 19 it is apparent that curves
corresponding to FSK2, FSK4 and MSK signals fall
below the chosen threshold level tr(γmax) = 4. This
threshold level remains the same for signals with
various parameters (symbol rate, carrier frequency,
sampling rate).
Then the spectrum analysis was used to
discriminate between the FSK2, FSK4 and MSK Fig. 21. Power spectrum of FSK2
modulations and instantaneous phase analysis to
discriminate between the ASK, BPSK, QPSK,
8PSK and QAM-16 modulations. The power
spectrum of MSK signal has only one carrier
frequency, the spectrum of FSK2 signal has two
maxima, which correspond to two carrier
frequencies, and the spectrum of FSK4 signal has
four maxima which correspond to four carrier
frequencies. The power spectrums of MSK, FSK2
and FSK4 signals are shown in fig. 20, 21, and 22.

Fig. 22. Power spectrum of FSK4

The centered non-linear component of the

instantaneous phase ϕ N L was analyzed to enable the

ISSN: 1109-2742 569 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

discrimination of ASK, MPSK and QAM-16

modulations.

2πf c i
ϕ NL (i ) = ϕ uw (i ) − , (12)
fs
where ϕuw(i) is the unwrapped phase sequence,
2πf c i
is the linear component of the instantaneous
fs
phase.
The centered non-linear components of the
instantaneous phase ϕNL for each modulated signal
are shown in fig. 23 - 26.
Fig. 25. ϕNL for 8PSK

Fig. 23. ϕNL for BPSK

Fig. 26. ϕNL for QAM-16

The BPSK signal has two phase values, QPSK

has four phase values, 8PSK has eight phase values
and QAM-16 has twelve phase values. The ASK
signal has only one phase value. For the analysis of
instantaneous phases, their histograms were
calculated. One, two, four, eight and twelve maxima
occur in phase histograms, which correspond to the
number of phase values in the signals.

4 Examination of Recognition and

Fig. 24. ϕNL for QPSK
Detection Algorithms
In this chapter, a method will be described that
we plan to use for the examination of algorithms
described above with real signals. The measuring
system will consist of a modulated signal generator,
receiver, digitizer and computer (fig. 27.).

ISSN: 1109-2742 570 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

sufficient. The necessary sampling frequency is

Generator Receiver Digitizer 500 kHz / 6 * 200 = 16.7 MHz.
The following instruments will be used in the
testing: signal generator Rohde&Schwarz
Fig. 27. System for practical verification of
SMIQ02B, receiver AOR SR2200, oscilloscope
recognition method
Agilent DSO6102A.
The generator provides ASK, MPSK, MFSK,
QAM, and MSK signals. A generator type with the
possibility of adding noise and simulating multipath
5 Conclusion
In this paper several algorithms for digital
signal propagation is preferred. In this case it is
modulation recognition and detection of symbol rate
possible to connect the generator and the receiver were described. The algorithms designed have good
directly by a cable. Otherwise it is advisable to use results when simulated on computer. However, in
antennas and let the signal propagate in a real the future work it is necessary to verify how these
environment. algorithms will work in practice. Currently we
The carrier frequency of modulated signal is assemble a measuring system which will test the
hundreds of MHz or units of GHz as is usual in designed algorithms with real signals.
common communication systems with digital
modulations. It is not easy to digitize signals at such
high frequencies and thus the generated signal is Acknowledgment
downconverted in the receiver. The receiver output This work was supported by the Ministry of
signal at an intermediate frequency (IF) of 10.7 Education of the Czech Republic research project
MHz is then digitized. For this purpose we use a No. MSM0021630513 and by the Czech Science
storage oscilloscope, but a digitizer with suitable Foundation project No. GP102/09/P626.
sample rate (it will be analyzed in the next
paragraph) can also be used. After digitization, the References:
signal is transferred into the computer where [1] W. Dai, Y. Wang, J. Wang, “Joint power
recognition is performed. estimation and modulation classification using
While designing the practical testing system it second- and higher statistics,” WCNC 2002 -
was necessary to determine the minimum sampling IEEE Wireless Communications and
frequency for signal digitization. The carrier Networking Conference, no. 1, 2002, pp. 767 –
frequency of the signal before digitization is 10.7 770.
MHz, which corresponds to the receiver IF. [2] L. Hong, K. C. Ho, “Identification of digital
Analyses showed that the carrier frequency must be modulation types using the wavelet transform”,
MILCOM 1999 - IEEE Military
at least four times higher than the symbol rate. It
Communications Conference, no. 1, 1999, 427
means that the maximum symbol rate can be 10.7
– 431.
MHz / 4 = 2.675 MHz. We have also found out that
[3] A. Kubankova, D. Kubanek, “Automatic
minimum number of samples in one symbol is Ns,min Digital Modulation Recognition Algorithm
= 200. It appears from this that the minimum Based on Spectrogram Analysis”, TSP 2008 -
sampling frequency at a maximum symbol rate of Telecommunication and signal processing
2.675 MHz is 2.675 MHz * 200 = 535 MHz. 2008 31th International Conference, 2008, s.
Another condition for a correct modulation 1-4.
recognition is a sufficient frequency bandwidth of [4] A. Kubankova, “Design and Analysis of New
the output IF filter in the receiver. It was found that Digital Modulation classification method”,
the pass band must be at least 6-times wider than the WSEAS Transactions on Communications,
symbol rate. For the above-mentioned symbol rate 2009, roč. 8, č. 7, s. 628-637. ISSN: 1109-
2.675 MHz the bandwidth would be 2,675 MHz * 6 2742.
= 16.05 MHz, which is unfortunately unavailable in [5] A. Kubankova, D. Kubanek, “Extended
usual receivers. They mostly offer a maximum method of digital modulation recognition”,
bandwidth of 500 kHz and thus it is possible to TSP 2010 - Telecommunication and signal
process signals with symbol rates of up to 500 kHz / processing 2010, 33th International
6 = 83.3 kHz in this case. But for the first rough Conference, 2010.
verification of the recognition method it is

ISSN: 1109-2742 571 Issue 9, Volume 9, September 2010

 WSEAS TRANSACTIONS on COMMUNICATIONS Anna Kubankova, David Kubanek

[6] A. Kubankova, “Detection of Signal Element

Length for Digital Modulation Classification”,
RTT 2008 - Research in Telecommunication
Technology 2008 9th International
Conference. Vyhne, Slovak Republic, 2008, s.
1-3.
[7] A. Kubankova, “Analysis of Symbol Rate
Detection method based on Cepstrum
Analysis”, TSP' 2009 - 32nd International
Conference on Telecommunications and Signal
Processing. Hungary: Asszisztencia Szervezo
Kft., 2009, s. 5-8.
[8] E. E. Azzouz, A. K. Nandi, “Automatic
Modulation Recognition of Communication
Signals,” Kluwer Academic Publishers,
Netherlands, 1996.
[9] D. Krstic, M. Stefanovic, N. Kapacinovic, S.
Jovkovic, D. Stefanovic, Probability Density
Function of M-ary FSK Signal in the Presence
of Noise, Interference and Fading, WSEAS
Transactions on Communications, 2008, Issue
5, Vol. 7.
[10] G. Kandus, M. Mohorcic, M. Smolnikar, E.
Leitgeb, T. Javornik, A Channel Model of
Atmospheric Impairment for the Design of
Adaptive Coding and Modulation in
Stratospheric Communication, WSEAS
Transactions on Communications, 2008, Issue
4, Vol. 7.
[11] M. Mahmoud, M. Raid, OFDM System under
the Combined Effect Analysis of the Phase
Noise, Carrier Frequency Offset, Doppler
Spread, and Amplifier Nonlinearity of the SEL,
WSEAS Transactions on Communications,
2008, Issue 7, Vol. 7.

ISSN: 1109-2742 572 Issue 9, Volume 9, September 2010

View publication stats