Unsupervised frequency clustering algorithm for null space estimation in wideband spectrum sharing networks

TL;DR

This paper introduces an unsupervised frequency clustering algorithm that efficiently estimates null space matrices in wideband spectrum sharing networks by exploiting channel correlations, reducing computational complexity.

Contribution

It presents a novel clustering method that forms frequency bin clusters with correlated channels, eliminating the need for prior channel knowledge and reducing eigenvalue decompositions.

Findings

Reduces computational complexity in null space estimation

Effectively clusters frequency bins with correlated channels

Significantly decreases number of eigenvalue decompositions needed

Abstract

In spectrum sharing networks, a base station (BS) needs to mitigate the interference to users associated with other coexisting network in the same band. The BS can achieve this by transmitting its downlink signal in the null space of channels to such users. However, under a wideband scenario, the BS needs to estimate null space matrices using the received signal from such non-cooperative users in each frequency bin where the users are active. To reduce the computational complexity of this operation, we propose a frequency clustering algorithm that exploits the channel correlations among adjacent frequency bins. The proposed algorithm forms clusters of frequency bins with correlated channel vectors without prior knowledge of the channels and obtains a single null space matrix for each cluster. We show that the number of matrices and the number of eigenvalue decompositions required to obtain the null space significantly reduce using the proposed clustering algorithm.