FIELD OF INVENTION
The present invention relates to designing a blind modulation classification (BMC)
method for single carrier modulated signals. This method classifies the modulation
schemes without having any prior knowledge of the signal parameters such as
intermediate carrier frequency, symbol rate, carrier phase, frequency offset and timing
offset. Here, we have considered a six class problem which includes binary phase shift
keying (BPSK), quadrature phase shift keying (QPSK), offset-QPSK (OQPSK), PI/4-
QPSK, and minimum shift keying (MSK), 16-quadrature amplitude modulation (16-
QAM). The novelty of this BMC is the inclusion of more number of circular modulations
in the classification problem which are very difficult to classify.
BACKGROUND OF THE INVENTION
BMC is a rapidly evolving area of signal exploitation with applications in signal
monitoring, spectrum management in cognitive radio, cooperative communication and
adaptive communication systems such as long term evolution-advanced (LTE-
advanced). BMC method is also known as automated modulation classification method.
The blind method reduces the processing time and increases the spectral efficiency of
the communication system due to the absence of redundant data or training sequence
in the transmitted signal for the purpose of parameter estimation. In fact, training
sequence-based or data aided approaches do not work well in the presence of carrier
frequency offset, timing offset and phase jitters or in high speed communication where
the property of training sequence is lost due to abrupt channel condition. On the other
hand, bandwidth efficient transmission not only increases the effective data rate of the
system, but also reduces the cost of spectral licensing.
In the past few years, the interest in BMC methods has increased with the emergence
of new communication technologies. BMC is a very sensitive part in the communication
process. The success rate of BMC should be very high for a successful communication.

However, the common approaches of BMC is the classification of modulation schemes
based on the knowledge of signal parameters such as carrier frequency, symbol rate,
carrier phase, frequency offset and timing offset. For instance, although the coherent
maximum likelihood (ML) modulation classification method has very good performance,
it requires the knowledge of all the signal parameters before entering into the
modulation classification process as discussed in W. Wei and J. Mendel, “Maximum-
likelihood classification for digital amplitude phase modulations,” IEEE Transactions on
Communications, vol. 48, no. 2, pp.189-193, Feb 2000 and J. Sills, “Maximum-likelihood
modulation classification for PSK/QAM,” IEEE Conference on Military Communication
Conference Proceedings (MILCOM), vol.1,1999, pp. 217-220. The ML estimation method
is having high computational complexity and also very sensitive with timing offset,
frequency offset, and other synchronization errors. It requires data acquisition,
preprocessing, and decision making which requires a partial knowledge of signal
parameters. Many other techniques for BMC such as, pattern recognition, 2D histogram
and many more discussed in A. Nandi and E. Azzouz, “Methods for automatic
modulation recognition of communication signals,” IEEE Transactions on
Communications, vol. 46, no. 4, pp. 431-436, Apr 1998 and A. Polydoros and K. Kim,
“On the detection and classification of quadrature digital modulations in broad-band
noise,” IEEE Transactions on Communications, vol. 38, no. 8, pp. 1199-1211, Aug 1990.
The 2D histogram method has a problem of storage and estimating 2D probability
density function (PDF) is also a difficult problem. There are various feature based (FB)
statistical methods which are more interesting than the other methods discussed in O.
Dobre, A. Abdi, Y. Bar-Ness, and W. Su, “Survey of automatic modulation classification
techniques: classical approaches and new trends,” IET Communications, vol. 1, no. 2,
pp. 137-156, April 2007. The hierarchical cumulant approach provided in A. Swami and
B. Sadler, “Hierarchical digital modulation classification using cumulants,” IEEE
Transactions on Communications, vol. 48, no. 3, pp. 416-429, Mar 2000, has significant
performance over AWGN. The cyclic cumulant approaches and their combined one
provided in W. Gardner and C. Spooner, “Signal interception: performance advantages

of cyclic feature detectors,” IEEE Transactions on Communications, vol. 40, no. 1, pp.
149-159, Jan 1992, and S. Soliman and S.-Z. Hsue, “Signal classification using statistical
moments,” IEEE Transactions on Communications, vol. 40, no. 5, pp. 908-916, May
1992, also has a satisfactory performance. The method based on elementary cumulant
does not work well over a fading channel and is not able to classify all the modulation
formats given in this classification problem because some features are common
especially in circular constellation QPSK, OQPSK, PI/4-QPSK and MSK.
SUMMARY OF THE INVENTION
Performing BMC of a received signal is an intermediary stage between signal detection
and demodulation process. It is the main task of a smart/adaptive transceiver. The BMC
is used to identify the modulation format of the received signal with no perception of
the transmitted data and many unidentified parameters at the receiver side, i.e., timing
and frequency synchronization errors and multipath fading channel. Here, we
contemplate the issue of BMC in the presence of a flat fading channel. Many methods
have been proposed based on ML and FB, i.e., higher-order elementary cumulants and
cyclic cumulants of the received signal. The solution offered by the ML method
experiences computational complexity and requires high signal to noise ratio (SNR).
Higher order elementary cumulants work well in AWGN channel, but perform
undesirably in fading channel environment. Here, the BMC method has been proposed
and implemented for linearly modulated signals for single carrier communication
systems. The method is FB and is combination of both elementary and cyclic cumulant
which gives optimal performance even at low SNR.
The objective of the proposed BMC method is that it performs even in flat fading
channel as the elementary cumulant doesn’t work well in fading channel. It also
reduces the processing time as it needs only a short burst of data, i.e., around two
hundred to three hundred symbols are sufficient for successful blind classification. Even
though the availability of the spectrum is limited, huge amount of spectrum is used just

to transmit the redundant data or training sequences and an immense portion of the
spectrum is not utilized by the primary users. The most effective way to overcome this
is by designing a blind method at the receiver. The blind process increases the
spectrum efficiency of communication systems without using redundant data or training
sequences, thus it can be easily adopted for adaptive communication systems. It
increases the effective data rate or absolves the spectrum for other primary users to
reuse it.
The BMC method which has been proposed and implemented here is more robust than
any other modulation classification technique and works well even in flat fading
channel. The proposed BMC methods have been implemented on a NI testbed through
a software defined radio (SDR). The measurement performances have been obtained
over realistic scenarios and compared with the simulation studies. The performance of
the proposed BMC has been compared with the existing elementary cumulant method.
The measurements have been carried out in an indoor environment of Signal Processing
for Wireless Communication Laboratory, Indian Institute of Technology (IIT) Patna.
Preceding Work
1.) Apparatus and method for robust BPSK/QPSK blind modulation classification against
errors in satellite control system US 7697639 B2:
This invention relates to an apparatus and a method for robust Binary Phase Shift
Keying/Quadrature Phase Shift Keying (BPSK/QPSK), against errors in a satellite
communication system.
2.) Blind modulation classification apparatus for use in satellite communication system
and method there of US 20060115013 A1:
This invention provides a blind modulation classification apparatus having improved
performance in non-ideal communication environment having frequency error and

phase error, by reducing computational complexity and classifying QAM and showing
less sensitivity with the frequency and phase error.
3.) Asymptotically optimal modulation classification method for software defined radios
US 8750425 B1:
This invention is directed to a likelihood test based modulation classification, modulation
schemes including PSK4, PSK8, PSK16, QAM8, QAM16, QAM 32, and QAM 64.
4.) Detection of signal modulation format type WO 2006102572 A1:
Generally this invention relates to mobile phone technology, and more specifically to
detecting the modulation format type of a signal, i.e., Gaussian minimum shift keying
(GMSK) and 8PSK modulation formats
5.) Blind modulation detection US 8331498 B2:
Generally this invention relates to wireless communication receivers, and in particular to
improved blind modulation detection and includes modulation formats MSK, QPSK, 8
PSK, 16 QAM or 32 QAM.
BREIF DESCRIPTION OF THE ACCOMPANYING DRAWINGS:
Fig.1: shows the flowchart of LabVIEW and MATLAB programming.
Fig.2: shows the block diagram of RF vector signal generator
Fig.3: shows the block diagram of RF vector signal analyzer.
Fig.4: Macro modulation classification by elementanry 4th order cumulant for six-
class of modulations in flat fading channel.

Fig.5: Non-zero cyclic peak is appeared at 4 times of IF carrier frequency of
QPSK modulated signals.
Fig.6: Non-zero cyclic peak is appeared at 4 times of IF carrier frequency of
OQPSK modulated signals.
Fig.7: Non-zero cyclic peak is appeared at 4wc ws/2 of PI/4-QPSK modulated
signals.
Fig.8: Non-zero cyclic peak is appeared at 4wc _ 2ws of MSK modulated signals.
Fig.9: Non-zero cyclic peak is appeared at ws of QPSK modulated signals.
Fig.10: Non-zero cyclic peaks for 2nd order cyclic cumulant of OQPSK modulated
signal, peaks seen in fig. are due to noise present in the signal
Fig.11: Front Panel Diagram for Transmitter (generating PSKs, 16QAM modulation
signal).
Fig.12: Front Panel Diagram for Receiver (classifying BPSK).
Fig.13: Front Panel Diagram for Receiver (classifying QPSK).
Fig.14: Front Panel Diagram for Receiver (classifying PI/4QPSK).
Fig.15: Front Panel Diagram for Receiver (classifying OQPSK).
Fig.16: Front Panel Diagram for Receiver (classifying MSK).

Fig.17: Front Panel Diagram for Receiver (classifying 16QAM).
Fig.18: Hierarchial blind modulation classification using cyclic cumulants and
elementary cumulants.
Fig.19: shows the success rate for measurement (M) and simulation (S) results
for six- class of modulation classification problem.
Fig.20: shows the success rate of comparison between Proposed (P) and
elementary cumulant (EC) for six- class of modulation classification
problem.
EC method is not able to classify PI/4QPSK and also not able to perform 100%
classification even at high SNR.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION:
System Model and Problem Statement
In this study, we assume that the signal is cyclostationary process (CP). In fact, most of
the man-made signals are CP. We use IF signal and complex baseband signal for micro
and macro blind modulation classification. A digitized IF signal is obtained by the down
conversion of the RF signal. Rough knowledge of the RF and baseband signal is
obtained by another down conversion and filtering of the IF signals. The IF is obtained
through coarse and fine carrier frequency estimation. The coarse IF carrier frequency is
obtained by smoothing the power spectral density (PSD) of the signal. Fine IF carrier
frequency is obtained by using the fourth order cyclic cumulant in G. Giannakis and G.
Zhou, “Harmonics in multiplicative and additive noise: Parameter estimation using cyclic
statistics,” IEEE Transactions on Signal Processing, vol. 43, no. 9, pp. 2217-2221, Sep

1995.In this context, a real discrete-time cyclostationary signal has been considered.
Let us start with a continuous-time domain signal; the transmitted RF signal in the
presence of timing offset and frequency offset signal is given by

Where L is the number of symbols, α [1] is the lth symbol drawn from a six-class
modulation schemes, T is the symbol period,r is the timing offset, fc is the carrier
frequency and φ is the carrier phase noise. For a successful blind demodulation, the
above signal parameters need to be estimated through a blind process. It is assumed
that g(f) is the transmitter band limited filter and without loss of generality it has been
assumed that g (t) is a root raised cosine pulse. The above expression can be deduced
as real and imaginary:

The received signal over AWGN can be written as:


where v(t) is the complex mixing AWGN with two sided PSD and variance σ2v =N O /2
The signal y(t) is over sampled at the rate of P/T to yield the discrete-time signal.
Here,

is a cyclostationary signal for P > 4 as mention in L. Mazet and P. Loubaton, “Cyclic
correlation based symbol rate estimation,” in Conference Record of the Thirty-Third,
Asilomar Conference on Signals, Systems and Computers, vol. 2, Oct 1999, pp. 1008-
1012, where Ts is the sampling instance and N1 is the number of samples
corresponding to delay r. The oversampling factor is defined as
P = FS/ fs =T/Ts =N/Lwhere, FS is the receiver sampling rate, fs is the symbol rate,
N is the signal length, and r5 is the sampling period which is small enough for all
possible transmitted baud rates, so that the oversampled signal is free from the inter-
symbol interference (ISI). The above real signal can be represented in terms of
complex signal as:

where y [ n] is the Hilbert transform of y~[N] In the problem statement, given
A/samples {y[n]1 N ,we need to estimate the carrier frequency of IF signal and perform

modulation classification without having the knowledge of other signal parameters. The
carrier frequency estimator and modulation classifier need only a short burst of data,
i.e., only two hundred to three hundred symbols are sufficient to classify modulation
schemes accurately.
Once the carrier frequency of IF signal is obtained we can obtain a baseband signal as

where, w[n] is a low pass complex baseband noise. We assume that the frequency
offset fain the above signal is of -SCJHZ which is negligible.
Feature Based: A Novel Modulation Classification Approach
1.) Macro Classification
FB methods first need some features for data representation. The macro classification is
based on the threshold detection given in A. Swami and B. Sadler, “Hierarchical digital
modulation classification using cumulants,” IEEE Transactions on Communications, vol.
48, no. 3, pp. 416-429, Mar 2000.The macro classification is used to classify real,
circular and rectangular constellation points, but this is very sensitive to phase noise
and frequency offset. The macro classification may fail over severe channel condition.
We define second order cumulant as:

Similarly, 4th order cumulant is given as


In general, 4th order cumulant can be written as
cum(a, b, c, d) = E(abcd) - E(ab)E(cd) - E(ac)E(bd) - E(ad)E(bc)
From above equation, we can find C 40 ,C 41 ,C 42 in terms of the 2nd and 4th order
cumulant of the baseband signal as in Fig.4. In terms of sample average, the above
equation can be written as

where N is the total number of received samples. It is noted that the above cumulant
equations are sum of the signal and noise. In practice, we estimate the normalized
cumulants by

where,C21 = C21-C21,w and C21w is estimated by system noise floor within a
signal bandwidth. From this elementary cumulant we can easily classify BPSK and
16QAM as shown in Fig.4.

2.) Micro Classification
Cyclic statistics are used to estimate the cyclic frequencies and phases with the help of
cyclic spectrum. However; this method becomes inadequate over a fading channel
especially if the signals are from the circular constellation. So, we have proposed a
novel approach for BMC, i.e., we have combined both elementary cumulant and cyclic
cumulant and used a hierarchical approach.
a.) Micro Classification at IF Signal
The 4th order time varying cumulant of y[n] at time lag o can be expressed as in G.
Giannakis and G. Zhou, “Harmonics in multiplicative and additive noise: Parameter
estimation using cyclic statistics,” IEEE Transactions on Signal Processing, vol. 43, no.
9, pp. 2217-2221, Sep 1995 as

where, c s [n;0] = cr~ and c φ [n;0] = 1, so for m=1 we find
second-order cyclic cumulant as:

where, cr2 a = a2r + oi 2 , above equation is maximum at ak=1/P

The above equation is for linearly modulated signals, such as BPSK, QPSK, 8-PSK and
16-QAM as shown in Fig.9. But for OQPSK there is a shift of P/2 samples, thus the in-
phase signal can be written as


Now its second order cumulant is obtained as

For m = i, the second order cyclic cumulant for OQPSK can be found by adding

From the above equation, it can be seen that C[s21][ak;0] has non-zero peaks
at-1/P, 0, 1/P. But the peak at cyclic frequency of ak = 1/P is being omitted due to
the existence of offset (P/2) and carrying equal power in I and Q
components C [s21][ak ;0]^ =1/P = 0 . as shown in Fig.10 Now we can classify
modulation format as shown in fig.4.
Implementation and Measurement
Configuration of the test bed is performed by using one transmitter of NI PXIe-5673
and one receiver of NI PXIe-5663. Both are composed by SDR. Fig.1 shows the basic
flow chart of LabVIEW and MATLAB programming. The measurements have been
performed in the Signal processing for wireless communications laboratory, IIT Patna.

Transmitter
To measure the performance of the blind receiver, a user configurable transmitter PXIe-
5673 has been shown in Fig.2 which consists of PXI-5652 (Local oscillator), PXIe-5611
(RF up converter) and PXIe-5450 (16-bit arbitrary waveform generator). Once the
baseband signal is generated, the signal is first converted into baseband samples with
the help of PXIe-5450 and then it is up-converted using PXIe-5611 to transmit it
through the connected antenna. The baseband signal is generated through different
modulation schemes, oversampling factors and symbol rates. The transmit power and
carrier frequency of the signal can be easily changed. Fig.11 shows the front panel of
transmitter.
Receiver
The blind receiver is implemented on NI PXIe-5663 as shown in Fig.3 which consists of
PXIe-5652 which is used as a Local Oscillator (LO),PXIe-5601 (RF down converter) and
PXIe-5622 (16-bit IF digitizer). By design, the receiver has been separated into three
sections: RF signal acquisition, IF signal processing and baseband signal processing.
The RF signal is controlled by LABVIEW programming and the IF and baseband signal
are processed through MATLAB script.
Signal Acquisition
In the RF search process as in Fig.1, firstly a specified RF filter span is selected. The
received RF energy is then compared with a predefined threshold value. The predefined
threshold is acquired through a calibration process where the energy of the received
signal is calculated within a specified span in the absence of signal. If the received
signal energy is above the threshold value, the position of the peak is estimated, which
is the estimated RF carrier frequency denoted by £ and the signal is processed to
obtain the IF signal.

Alternatively, the next burst signal is captured and the same process is repeated. The
signal is down converted to IF by setting the local oscillator frequency as /c'-5MHz,
where the proposed method needs a sampling rate of eight times the IF carrier to
estimate the IF carrier without much ambiguity. Thus, an IF carrier is set at about 5MHz
and sampling rate of 50MHz.
IF and Baseband Signal Processing
The received digitized IF signal is passed to the MATLAB script as shown in fig.1 which
runs inside the LABVIEW environment. The received real IF signal y[n] is first
converted to complex IF signal y[n], and then IF carrier and signal bandwidth are
estimated from PSD of the complex IF signal. However, modulation classification is
performed both at IF and baseband level as shown in Fig.1 and Fig. 18. Fig. 12, 13, 14,
15, 16 and 17 shows the front panel of receiver classifying different modulation
schemes.
Simulation and Measurement Results
In this section, the performance of BMC method is evaluated through simulation and
measurement results as shown in Fig The simulation and experiment parameters are
provided in Table 1.



table 2 and 3 presented below shows the modulation classification through simulation
at 10dB & 15dB SNR and table 4 shows the modulation classification through
measurement at 10dB SNR.



CONCLUSION
In this section we have discussed about the brief idea of the work that was
accomplished in this work. The proposed method is based on combination of
elementary cumulant and cyclic cumulant. The extensive implementation and simulation
are carried out, and the proposed method outperforms than several feature based
methods. This approach is robust in the presence of synchronization errors, performs
even better at low SNR, as well as having a better classification performance in fading
channel.
REFERENCES
[1] S. Majhi and T. Ho, “Blind symbol-rate estimation and test bed implementation of
linearly modulated signals,” IEEE Transactions on, Vehicular Technology, vol. 64, no. 3,
pp. 954-963, March 2015.
[2] S. Majhi and W. Xiang, “Blind symbol rate estimation and testbed implementation
for linearly modulated signals,” 78th IEEE Conference on, (VTC Fall), in Vehicular
Technology, Sept 2013, pp. 1-5.

[3] A. Swami and B. Sadler, "Hierarchical digital modulation classification using
cumulants," IEEE Transactions on, Communications vol. 48, no. 3, pp. 416-429, Mar
2000.
[4] G. Giannakis and G. Zhou, “Harmonics in multiplicative and additive noise:
parameter estimation using cyclic statistics," IEEE Transactions on, Signal Processing,
vol. 43, no. 9, pp. 2217-2221, Sep 1995.
[5] http://www.ni.com/pxi/.
[6] O. Dobre, A. Abdi, Y. Bar-Ness, and W. Su, “Survey of automatic modulation
classification techniques: classical approaches and new trends,” IET, Communications,
vol. 1, no. 2, pp. 137-156, April 2007.
[7] J. Sills, “Maximum-likelihood modulation classification for psk/qam,” IEEE
Conference on, Military Communications Conference Proceedings, MILCOM, vol. 1,
1999, pp. 217-220 vol.1.
[8] W. Wei and J. Mendel, “Maximum-likelihood classification for digital amplitude phase
modulations," IEEE Transactions on, Communications, vol. 48, no. 2, pp. 189-193, Feb
2000.

WE CLAIM
1) A blind modulation classification (BMC) method for single carrier, linearly modulated
signals, characterized by, combination of elemental and cyclic cumulants,
comprising six-class modulation problems, i.e; BPSK, QPSK, OQPSK, PI/4-QPSK,
MSK, 16-QAM.
2) A method as claimed in claim (1) includes performance (BMC) without any
knowledge of intermediate carrier frequency of the signal.

3) A method as claimed in previous claims includes performance (BMC) without any
knowledge of symbol rate of the signal.
4) A method as claimed in previous claims includes performance (BMC) without any
knowledge of carrier phase of the signal.

5) A method as claimed in previous claims includes performance (BMC) without any
knowledge of frequency off-set of the signal.
6) A method as claimed in previous claims includes performance (BMC) without any
knowledge of timing off-set of the signal.
7) A method as claimed in claim (1), wherein, the system is adaptive for wireless
transceiver with increased spectral efficiency.
8) A method as claimed in claim (1), wherein the system is adaptive for blind
demodulation.
9) A method as claimed in claims (1) & (8), wherein, the system is adaptive at low
signal to noise ratio.