Novel Frequency Offset Estimation Scheme for Reliable Wireless Communication using Modified K-Means Clustering

This paper presents a novel machine learning-based algorithm to estimate the frequency offset in wireless communication systems from the received signal's IQ constellations. The algorithm focuses on dividing the received signal symbols into clusters and finding the centroid of each cluster using a modified k-means algorithm. The constellation rotation angle, corresponding to the frequency offset, is then found from the angel between the obtained centroids' coordinates and the coordinates of the corresponding modulation format. The simulation results have shown 100% estimation accuracy for constellation rotation angles within the (-45, 45) degrees range. The proposed algorithm provides a low-complexity scheme that eliminates the overhead required for training preamble (TP) based techniques, thus improving the communication system's efficiency. The algorithm can be applied to wide range of communication systems especially those used in high speed train and vehicular communications.