Blind Signal Digital Modulation Classification through k-medoids Clustering

Abstract

Modulation scheme classifier for the received RF signal is proposed in this paper. Modulation classification is an intermediate step between data detection and its demodulation for extracting the final information. Proposed method estimates the carrier frequency offset after downconversion of passband signal with estimated carrier frequency and corrects for it. Signal is sampled with high frequency and symbol rate is estimated. Sampled signal is downsampled to estimated symbol rate to extract the constellation points. For identification of modulation scheme between QAM and PSK of different orders, k-medoids clustering is used. Blindly, k medoids are estimated for k equals to 4, 8, 16 and 64 and similarity to ideal constellation structure is calculated. Final decision for modulation scheme is given in favor for which similarity with ideal constellation structure is maximum. The simulation results for the method shows that different modulation schemes are classified efficiently above 10 dB SNR in presence of Additive White Gaussian Noise. Method proposed is unsupervised and has low computational complexity.