Sphere decoding, which has been presented as a simplified method of ML decoding of lattice codes and ML detection of MIMO systems, achieves near-ML performance with relatively low complexity. However, improper choice of the sphere radius can cause either the decoding failure or increment of the computational complexity.
In this thesis, we propose the combination of the V-BLAST detection and the sphere decoding. The new sphere decoder does not exploit the probabilistic characteristics of the MIMO systems to determine the sphere radius, and prevents the occurrence of the decoding failure, and achieves the exact ML performance. And, since the sphere radius of the new sphere decoder is very tight even in the bad channel and the optimal ordering of V-BLAST detection reduces the span of the search space of lattice points, the new sphere decoder is less computationally complex than the original sphere decoder.
We also present various simulations on the complexity and performance of the new sphere decoder and show that our new sphere decoder achieves a considerable complexity reduction and an improved performance compared with the original sphere decoder.