Modulation classification of linearly modulated signals in a cognitive radio network using constellation shape

In this paper we intend to classify linear digital modulations in a cognitive radio network using constellation shape of the received signal as a feature. In our method, we perform the clustering of the base-band symbols to recognize the constellation and evaluate the result by some validation method. For clustering, K-Means, one of the simplest and fastest clustering methods, and the Fuzzy Hyper Volume (FHV), a well-known, simple and fast index for validating fuzzy C-Means clustering method, are employed. The presented approach is fully unsupervised and performs its task despite the slowly-varying flat Rayleigh fading channel and without the knowledge of the parameters such as SNR, carrier phase and timing offset. Using this fast and accurate pair of method-index and introducing a novel idea for refining K-Means initial values that considerably increases accuracy and decreases the number of iterations required for K-Means clustering, guarantees high performance and low computational complexity of the aforementioned algorithm. It is shown via simulations that the algorithm performs correct classification when the SNR is over 10 dB for most practical linear modulation schemes under the presence of AWGN.