| This paper proposes a new signal detection algorithm for multiple-input multiple-output (MIMO) wireless communications. The optimal signal detection for these communications is the maximum likelihood detection (MLD), which is based on the maximum likelihood estimation. However, the computational complexity of MLD becomes prohibitive for implementation, when the number of transmit antennas and the modulation order are large. Thus, suboptimal detection algorithms that can reduce the complexity while maintaining almost the same bit error rate (BER) performance as MLD are required. The proposed algorithm sets a minimum mean-square error (MMSE) detection result to a starting point, and searches for signal candidates in multi-dimensions of the noise enhancement from which the MMSE detection suffers. The MMSE detection is classified as linear detection and requires a very small amount of complexity. When the channel matrix is ill-conditioned, however, the BER performance of the MMSE detection degrades seriously due to the noise enhancement. Therefore, the transmit signal is very likely to exist nearly in the directions of the noise enhancement with the MMSE detection result being a starting point, which the proposed algorithm exploits. In highly spatially correlated MIMO channels such as mobile communications, there are plural dominant directions of the noise enhancement, and thus multi-dimensional search is necessary. To reduce the complexity of the multi-dimensional search, the proposed algorithm limits the number of signal candidates to the order of the number of transmit antennas. Specifically, the signal candidates, which are unquantized, are obtained as the solutions to minimize the log likelihood function under a constraint that a stream of the candidate should be a constellation point. Finally, the detected signal is selected from hard decisions of both the MMSE detection result and the unquantized signal candidates on the basis of the log likelihood function. In addition, the proposed algorithm is applicable to coded MIMO systems, because it can easily calculate log likelihood ratios of coded bits that a channel decoder requires. Computer simulations under a correlated MIMO channel condition demonstrate that the proposed scheme is superior to a conventional one-dimensional search algorithm in BER performance while requiring almost the same amount of complexity as that of the conventional algorithm. In summary, the proposed signal detection algorithm can achieve excellent BER performance even in highly correlated MIMO channels, while reducing the complexity to the amount suitable for implementation. |
