A probabilistic approach to Integer Least Squares and MIMO decoding
In this work we propose a novel probabilistic approach to solving the integer least squares (ILS) problem, where the unknown variables are constrained to be in a finite set. First, we introduce a novel subspace projection operator that estimates the pairwise assignment probabilities of the ILS variables. Thus, we cast the ILS as a probabilistic inference problem. Second, we formulate the ILS inference as a pairwise assignment problem and propose a computationally efficient spectral approach to its solution. We demonstrate our approach by applying it to MIMO decoding, a problem that attracted significant research interest. This provides a viable set of baseline algorithms to compare against. Under a wide range of conditions our method compares favorably with contemporary state-of-the-art MIMO decoding schemes.