A Bayes Decision Approach to Enhance MIMO Wireless System Performance

The MIMO system takes advantage of the spatial diversity gain using spatially separated antennas for both the receiver and transmitter. This effectively mitigates the fading effects and increases the channel capacity in rich Rayleigh multipath environments [1, 2, 3]. In theory, MIMO signals propagate over an independently and identically distributed (i.i.d.) multipath fading channel that results in a linearly increasing channel capacity with the minimum number of transmission and receiving antennas. However, in reality, the inter-subchannel correlations and channel gain imbalances due to inadequate scattering and/or inadequate antenna spacing (i.e. spatial correlations between antennas in both Tx/Rx antennas) causes signal dependent interference, the so-called coantenna interference (CAI) which is critical to the system capacity and power efficiency. In this paper, we propose an M-by-N multiple hypotheses Bayes decision rule (BDR) [4, 5] for the channel covariance optimization that is considered under the imbalance channel gain between MIMO channel coefficients [3] without the channel knowledge to the transmitter. Thus, the Mlikelihood optimum decisions for the antenna received signals is determined by multiple inequalities as functions of the conditional probability density function of the channel coefficients, cost factors associated with the decision, and a priori probabilities of the channel coefficients between receiver and transmitter antennas. With our soft-decision rule, the inverse values of the square root of the channel covariance are applied for the cost factors. This scheme effectively increases the mean signal-to-noise ratio (SNR) invoked with an achieved optimum channel covariance. It is furthermore demonstrated that the cumulative distribution functions (cdf) of the eigenvalues generated with the BDR is considerably better than that of the original one when the imbalance power ratio occurs over MIMO spatially correlated sub-channels.