Decoding algorithms in mainstream brain-computer interface systems (such as Utah arrays and ECoG electrode grids) mostly rely on single-electrode firing rates or power spectral features of local field potentials, and make little use of cross-electrode spatiotemporal phase relationships.