Compensation of Phase Noise in 5G NR with Machine Learning

The growth in mobile data traffic due to the immense increase in the number of connected personal devices requires a next generation of mobile communications capable of providing new services and applications. The fifth generation 5G New Radio (NR) is the one implemented this decade, with the objective of improving user experience and connecting society in an intelligent environment. One of the key technologies to meet the demands of higher system capacity is the use of Ultra-Dense Networks (UDNs) in millimeter-wave (mmW) frequencies. However, the combination of higher frequencies and the OFDM waveform may increase the system sensitivity to phase noise (PN). In this work, we compensate PN by using classical pilot-based estimation and extend this method by using machine learning (ML) algorithms. Both Gaussian mixture model (GMM) and weighted k-nearest neighbors (KNN) techniques are proposed. We show that with them a reduction of the required amount of pilots is achieved, while maintaining an affordable complexity.