Efficient MIMO detection by successive projection

Abstract

This paper presents a new MIMO detector based on successive projection of the received signal onto the faces of the lattice induced by the channel matrix. Analysis of the relationship between successive algorithms like V-BLAST and the maximum likelihood (ML) sphere decoder, which share a common underlying tree structure, leads to the design of the successive projection algorithm (SPA). Although it is a suboptimal detector, we prove theoretically that when ML detection can be realized in a successive manner, i.e., without back-tracking, ML performance is also achieved by the SPA. The advantageous implications of this result are illustrated via simulation of the average bit error rates attained over a 4 times 4 MIMO flat fading channel. For instance, at a target error rate of 10-3 using 16-QAM modulation, a 2.8 dB improvement over the popular V-BLAST detector is observed at a comparable complexity. We also demonstrate a parameterized SPA, which offers performance profiles approaching that of an ML detector