"Automatic Modulation Classification (AMC) is a process which classifies the modulation scheme of a
received signal. It is followed by demodulation process for accurately retrieving the information present
in the received signal. A prominent application of AMC is in military intelligence to decode a signal of
interest. Also, an energy efficient jamming signal can be designed after knowing the specific modulation
scheme. In civil applications, AMC is used in software defined radio and in cognitive radio to design
single receiver for different type of transmitters to select appropriate demodulation approach.
Generally, there are two groups of algorithms for classification, one is Likelihood Based (LB) Algorithms
and second is Feature Based (FB) Algorithms. LB algorithms give optimal solutions but have
high computational complexity, whereas FB algorithms are of reduced complexity but give a less optimal
solution. With new techniques, FB algorithms can give near optimal solution for properly chosen
features. Features used in FB algorithms are Instantaneous features like amplitude, frequency, phase;
Transform based features such as Fourier and Wavelet transform; Higher order moment, cumulants
based and Cyclostationarity based.
In the proposed work, analytic Morlet wavelet is used as mother wavelet to extract features
presented in [Sethi and Ray, 2013] which were originally extracted from the baseband signal. These features
are further used as the input to different machine learning classifiers viz. Support Vector Machine,
KNN and Bagged Tree, and results are compared to the baseline model in which Haar wavelet was used
to extract four features from four type of signals to get a set of sixteen features. The same set of sixteen
features is extracted from Daubechies mother wavelet. Average accuracy obtained for Morlet wavelet
is 96.48%
Keywords: AMC, Wavelet Transform, Morlet, SVM, KNN, Bagged Tree
i"