STATISTICAL-ANALYSIS OF DECENTRALIZED MUSIC

For large-array processing problems there is a need for decentralized methods that are computationally efficient. A decentralized variant of the MUSIC algorithm, proposed in the literature, is here analysed from a statistical viewpoint, assuming that a large number of snapshots is available. As intuitively expected, the centralized MUSIC algorithm is more accurate than both the local and decentralized versions of MUSIC. More surprisingly, the decentralized variant of MUSIC is not always more accurate than the local MUSIC estimates. The theoretical analysis is supported by numerical examples.