Optimal Projection Method in Sphere Decoding
Arash Ghasemmehdi, Erik Agrell · arXiv (Cornell University) · 2009
An entirely different approach to complexity reduction in sphere decoders is taken. Here we demonstrate that most of the calculations in the standard algorithms are in fact redundant in the sense that the calculated values are never used. This applies to all recursive sphere decoder algorithms, including the numerous variations of the Fincke-Pohst and Schnorr-Euchner strategies. We propose a method, which is applicable to lattices as well as finite constellations, to avoid these redundant calculations, thus reducing the complexity. We emphasize that the algorithms otherwise perform exactly as before, visiting the same points in the same order, and returning the same result. Pseudocode is given to facilitate immediate implementation. In simulation results, it is shown that the relative complexity gain with the proposed add-on goes up linearly as the dimension of the lattice increases. For instance, the complexity is reduced to one fourth for lattices at dimension sixty.