Soft Information in Interference Cancellation Based Multiuser Detection

This paper considers soft information in interference cancellation based multiuser detection algorithms. For the case of binary phase-shift keying (BPSK) modulation, a relationship between the soft information as provided by the log likelihood ratio (LLR) and the minimum mean-square error (MMSE) bit estimates is derived, whereby the bit estimate is given by the hyperbolic tangent of one-half the LLR. Although this result is known for Gaussian channels, we show that this result holds true regardless of the underlying channel distribution. Similar relationships are derived for quadrature phase-shift keying (QPSK) and also for more general constellations. Results for multiuser detection (MUD) are then derived. Starting with an analysis of the convergence dynamics, what follows is an analysis of four canonical iterative MUD algorithms that fit within the framework: parallel vs. serial interference cancellation and symbol-level vs. chip-level updates. The results help explain why chip-level algorithms lead to faster convergence for a given amount of computation than do symbol-level algorithms, even though the chip-level algorithms never explicitly calculate the matched filter that is known to be a sufficient statistic. Finally, an efficient computational structure for a MUD processing element for the chip-level parallel interference cancellation (CPIC) is derived as a one-chip ahead prediction.