Extrinsic versus a Posteriori Probability based iterative LMMSE-IC algorithms for coded MIMO communications: Performance and analysis

Several studies dealing with iterative Linear Minimum Mean Square Error Interference Cancellation (LMMSEIC) algorithms have concluded that better performances are obtained when soft interference cancellation is performed using Log a Posteriori Probability Ratios (LAPPRs) on coded bits rather than Log Extrinsic Probability Ratios (LEXTPRs) on coded bits. While LMMSE-IC can be rigorously derived and analyzed when it is based on LEXTPRs, things become more complicated when it relies on LAPPRs. The aim of this paper is twofold: (1) Emphasize the underlying assumptions needed for the derivation of (iterative) LAPPR based LMMSE-IC, (2) Propose an improved performance prediction method (or physical layer abstraction) robust when these assumptions do not hold. The latter contribution is a cornerstone for cross-layer design and optimization in presence of advanced turbo receivers in 3GPP.