An Enhanced Deterministic Sequential Monte Carlo Method for Near-Optimal MIMO Demodulation With QAM Constellations

We propose a low-complexity, linear minimum mean squared error (MMSE)-based sequential Monte Carlo (SMC) technique as an alternative to the sphere decoder for near-optimal demodulation in multiple-input multiple-output (MIMO) systems. Prior to the SMC procedure, the received signal is passed through a linear MMSE-based preprocessing step, which also determines an optimal channel-dependent order of detection and produces a sequential structure. The algorithm then draws the symbol samples in a deterministic fashion, and the survivor paths are selected based on their importance weights. The proposed algorithm exploits the rectangular structure of the QAM signal constellation by separating the real and imaginary parts of the signal to reduce the complexity associated with the listing and weight update steps, resulting in a complexity (in terms of the constellation size $M$ ) of ${\cal O}(\sqrt{M})$ as compared to ${\cal O}(M)$ complexity of the existing SMC algorithms for an $M$-QAM constellation. We demonstrate through simulations that the new method achieves the sphere decoder performance for V-BLAST systems. Unlike the sphere decoder whose complexity is channel-dependent, our algorithm has a fixed complexity which is channel independent; thus it is well suited for use in practical MIMO systems. Some other interesting features of the algorithm are that it is able to handle MIMO systems with less receive antennas than transmit antennas; and can also deal with multiuser multirate MIMO systems, utilizing a novel ordering scheme. Finally we extend the proposed algorithm to solve the lattice decoding problem and demonstrate the effect of different preprocessing stages on the performance and complexity of the algorithm through extensive simulation results.