Bit-Level Reduced Neighborhood Search for Low-Complexity Detection in Large MIMO Systems

Neighborhood search algorithms (NSAs) reduce the detection complexity in a multiple-input multiple-output (MIMO) system equipped with a large number of antennas. These algorithms iteratively search for the optimal maximal likelihood (ML) vector in a chosen neighborhood and therefore, their performance depends on the probability that the desired solution vector belongs to the neighborhood. An efficient choice of neighborhood vectors which are likely to reduce the ML cost, can in-turn reduce the complexity of the NSAs. To enable this, we propose a novel MIMO detection framework by representing the transmit symbols as a polynomial function of its constituent bits. We use this framework to propose: 1) a bit-level extension for the minimum mean squared error detector to initialize neighborhood search and 2) a metric-based selection criteria to reduce the neighborhood size. Combining the two ideas, we re-frame the NSAs, namely, likelihood ascent search and reactive tabu search, and numerically show that the proposed approach significantly reduces the complexity without affecting the bit error rate.