Path-based spectral clustering for decoding fast time-varying MIMO channels

In this paper, we present a clustering technique for decoding fast time-varying multiple-input multiple-output (MIMO) channels. The proposed method builds upon previous work that exploited the symmetry of the constellation and the order of the data within a spectral clustering procedure. The novelty of this work is that by adjusting the different steps of the standard spectral clustering algorithm, it introduces the expected shape of the clusters into the clustering process. The main modification applies to the construction of the weighted graph, for which it is shown that a path-based kernel, the connectivity kernel, can be a more appropriate similarity function than the Gaussian kernel. The obtained spectral clustering method is capable of finding clusters in sequential data. Experimental results are included to demonstrate the validity of the method.