Bridging the Gap Between S-PARC and Optimum Closed-Loop MIMO

An optimum closed-loop multiple-input-multiple-output (MIMO) scheme requires a large amount of instantaneous channel information to be fed back from the receiver to the transmitter. Another closed-loop MIMO scheme called (selective) per antenna rate control (S-PARC) requires much less feedback. However, it does not perform well in conditions of low SNR, correlated antennas and Ricean fading where the number of independently coded and modulated data streams that can be supported by the channel, K, is less than the number of transmit antennas M. This is because S-PARC uses K out of M antennas to transmit the streams, which limits its capacity. The problem can be alleviated by using all antennas to transmit the streams. This is done in multiple stream co-phasing (MSC) by partitioning the set of M transmit antennas into K mutually exclusive subsets for transmission of K independent data streams after "co-phasing". Co-phasing can be considered as instantaneous beamforming where relative phase rotations (co-phasing angles) are applied to antennas for transmission of a stream. The co-phasing angles for each stream are designed to maximize the stream capacity (approximately). A minimum mean square error - successive interference cancellation (MMSE-SIC) receiver is used in MSC, as in S-PARC. MSC requires feedback of K rates and M-K angles, i.e. M real coefficients, same as in S-PARC. The average capacity of the various schemes is evaluated for 4times4 and 4times2 MIMO systems for flat Rayleigh and Ricean fading with antenna correlation. It is found that in the presence of fewer receive antennas than transmit antennas, low SNR, antenna correlation and Ricean fading, MSC greatly outperforms S-PARC. This is because MSC uses all antennas to transmit an adaptive number of streams after co-phasing, achieving high SNR. MSC closely approaches the capacity of optimum closed-loop MIMO for the entire SNR range for all channel models considered. Thus, the restriction of antenna set part- itioning greatly reduces the amount of feedback in MSC, as compared to optimum, without significantly compromising performance.