A Spectral Clustering-Based Pilot Assignment Method for Cell-Free Massive MIMO

This paper studies pilot assignment (PA) in cell-free (CF) massive multiple-input multiple-output (MIMO). PA is a key issue to facilitate signal processing, e.g. beamforming, by providing reliable channel estimation at access points (APs). Most existing methods face intractable local optima problem, which degrades the channel estimation accuracy. Moreover, these methods usually encounter the unaffordable complexity. To this end, we propose a low-complexity PA method based on the spectral clustering (SC), which is named as SC-based PA method. SC is a popular machine learning method for clustering problem in image segmentation, bioinformatics, etc. It leverages the eigenvectors of a similarity matrix to map the data samples to a low-dimensional space for easy clustering. By assigning the orthogonal pilot sequences to the same-cluster user equipments (UEs), the proposed SC-based PA method properly degrades the pilot contamination between the UEs. To verify the performance of the proposed method, we show by simulations that the proposed SC-based PA method outperforms the existing methods in both channel estimation accuracy and execution time.