MMSE Equalization-Based MIMO Demodulation with Random Walk along Gradient Descent Direction

This paper proposes a minimum mean square error (MMSE) equalization-based multi-input multi-output (MIMO) demodulation scheme with Metropolis-Hastings random walk along a gradient descent direction. The conventional scheme using intermediate search points works effectively for higherorder modulation symbols. However, its performance is deteriorated for lower-order modulation symbols. The proposed scheme suppresses the coefficients in the offdiagonal components of a channel matrix by applying MMSE equalization and improves performance in low signal-to-noise power ratio conditions even for lowerorder modulation symbols. Numerical results obtained thorough computer simulation show that the proposed scheme improves by about 2.0 dB at a bit error rate of $10^{-2}$ when the number of iteraitons is 10 for QPSK symbols with 64 transmit and receive antennas.