Coset-based lattice detection for MIMO systems

This paper presents a new sub-optimal lattice decoder based on decomposition of the detection lattice into cosets and iterative application of LLL reduction to its primal and dual representations. From each coset, a candidate solution is chosen using linear equalization followed by integer quantization. These candidates are then compared and the best amongst them is returned. Because the cosets are identical save for distinct translation vectors, their sub-lattice decoders share common preprocessing and can operate very efficiently in parallel. Exclusive of pre-processing, the computational complexity of the proposed decoder is quadratic in the problem dimension and linear in the number of cosets. Simulation results demonstrate the near-ML bit error rate performance offered by the new decoder in the context of existing QAM-modulated MIMO detection schemes.