Reducing sphere decoder complexity by elliptical tree pruning

We propose a computationally efficient detection algorithm with near ML performance for space-time communications systems. Our algorithm is based on pruning (excluding) branches from the search tree arising in the sphere decoding (SD) algorithm. The choice of the branches to be pruned relies on intuitive geometrical reasoning which is in turn based on a simplified channel model designed to capture the effect of badly conditioned channel realizations. The remaining search tree after pruning can be processed by the conventional SD algorithm with strongly reduced complexity as against scanning the complete tree. Simulation results show that despite the pruning near ML results can be obtained.